Defective portion output method and defective portion output device

Abstract

The controller (10) calculates a feature amount of target mesh data on the basis of nodes and connection lines constituting the target mesh data which is created by converting the three-dimensional shape data and which is the target of analysis processing, and inputs the feature amount of the target mesh data as input data into the learned model. And a learning model that is obtained by learning using the training data and that is used to output a site or position in the target grid data where there is a defect portion that needs to be corrected. The training data is obtained by associating a location or position in the learning grid data in which the defective portion has been identified with a feature amount of the learning grid data when the defective portion has occurred in the learning grid data.

Description

Technical Field

[0001] The invention relates to a defective part output method and a defective part output device. Background Art

[0002] The following technology is known: a three-dimensional model based on three-dimensional CAD (Computer Aided Design) data is meshed to generate a three-dimensional model for analysis, and the difference in volume, surface area and center of gravity between the three-dimensional model based on the three-dimensional CAD data and the three-dimensional model for analysis is calculated and displayed to the operator (Patent Document 1).

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Publication No. 2004-94663 Summary of the invention

[0006] Problem that the invention aims to solve

[0007] However, in the technology of Patent Document 1, the part to be corrected is not displayed in the three-dimensional model for analysis, so there is a problem that the operator needs to visually check the three-dimensional model for analysis to determine the part to be corrected, and the three-dimensional model for analysis cannot be corrected efficiently.

[0008] The problem to be solved by the present invention is to provide a defective portion output method and a defective portion output device capable of efficiently correcting a three-dimensional model for analysis.

[0009] Solutions for solving problems

[0010] The present invention solves the above-mentioned problem by the following manner: based on the nodes and links constituting the object mesh data which is produced by converting the three-dimensional shape data and is the object of the analysis process, the feature quantity of the object mesh data is calculated, the feature quantity of the object mesh data is input as input data into a learned model, and the learned model is used to output the part or position of the defective part in the object mesh data which needs to be corrected. The learned model is obtained by learning using training data, wherein the training data is the training data obtained by establishing an association between the part or position of the defective part in the learning mesh data in which the defective part has been determined and the feature quantity of the learning mesh data when the defective part is generated in the learning mesh data.

[0011] Effects of the Invention

[0012] According to the present invention, a three-dimensional model for analysis can be efficiently corrected. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a block diagram showing a schematic structure of a defect output device according to an embodiment of the present invention.

[0014] Figure 2 This is a diagram showing an example of application of the defective portion output device involved in this embodiment.

[0015] Figure 3 This is a diagram showing an example of an output result of a defective portion according to the present embodiment.

[0016] Figure 4 This is a flowchart showing an example of the process of the learning method of the machine learning model involved in this embodiment.

[0017] Figure 5 This is a flowchart showing an example of the process of the defective portion output method involved in this embodiment.

[0018] Figure 6 This is a flowchart showing an example of the process of the defective portion output method involved in this embodiment. DETAILED DESCRIPTION

[0019] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0020] Figure 1 1 is a block diagram showing the schematic structure of a defective part output device involved in one embodiment of the present invention. The defective part output device 1 is applied to a system for producing mesh data used in CAE (Computer Aided Engineering) analysis. In CAE analysis, structural mechanics, fluid mechanics, etc. are analyzed for mesh data produced by converting the design shape data of the analysis object. The design shape data of the analysis object is three-dimensional shape data produced by software such as CAD. Mesh data is data produced by meshing the three-dimensional shape data. That is, mesh data refers to data that expresses the design shape of the analysis object through a mesh. Mesh data includes data on nodes and lines that constitute the mesh. In addition, the node includes data on node numbers and coordinates for each node. The coordinates of the node are coordinates that represent the position of the node in three-dimensional space. The line data is data on the connection relationship between nodes. In each mesh, multiple nodes are connected in a manner that forms a triangle or a quadrilateral.

[0021] Mesh data is obtained by converting the three-dimensional shape data of the object to be analyzed by existing automatic mesh software functions. In the past, mesh data produced by existing automatic mesh software functions had defects that needed to be corrected. Therefore, a person needs to visually confirm the three-dimensional model based on the mesh data displayed on the display to determine and correct the defective part, which makes it impossible to correct efficiently. In this embodiment, the defective part output device 1 can efficiently correct the mesh data by automatically detecting and outputting the defective part of the mesh data.

[0022] use Figure 2 An example of applying the defective portion output device according to the present embodiment will be described. Figure 2 (a) shows an example of a three-dimensional model based on three-dimensional shape data to be analyzed by CAE. Figure 2 Three-dimensional shape data as shown in (a). Figure 2 (b) shows an example of a three-dimensional model based on mesh data converted from three-dimensional shape data. Figure 2 The mesh data is as shown in (b). Figure 2 (c) shows that Figure 2 The defective parts P1 and P2 generated in the three-dimensional model based on the mesh data shown in (b) of FIG. The defective part output device 1 automatically detects defective parts such as the defective parts P1 and P2 and outputs them to the operator.

[0023] Defective parts generated in mesh data are generated in large quantities in the analysis of a structure formed by assembling multiple parts, such as an automobile. Since it is an assembled structure, there are a large number of fillets and welds, resulting in a large number of defective parts in mesh data. Defective parts generated in mesh data are, for example, defects in which a part that was originally an edge line in the three-dimensional shape data does not become an edge line, defects in which a part that was originally a fillet or a weld bead in the three-dimensional shape data is out of shape, and defects in the confluence.

[0024] The defective part output device 1 includes a controller 10 and a database 2. The controller 10 includes a computer having hardware and software, and the computer includes: a ROM (Read Only Memory) storing a program; a CPU (Central Processing Unit) executing the program stored in the ROM; and a RAM (Random Access Memory) functioning as an accessible storage device. In addition, as an action circuit, an MPU (Micro Processing Unit), a DSP (Digital Signal Processing), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), etc. can be used instead of or in conjunction with the CPU.

[0025] The controller 10 estimates the location or position of the defective part of the mesh data using the learned model, thereby detecting the defective part, and outputting the detected defective part. The controller 10 is configured to include a data input unit 11, a feature quantity calculation unit 12, a learning unit 13, and an output unit 14 as functional blocks, and performs each function through the cooperation of hardware and software for realizing the above-mentioned functions or executing each process. In addition, in this embodiment, the functions of each functional block are explained on the basis of dividing the functions of the controller 10 into four blocks, but the functions of the controller 10 do not necessarily need to be divided into four blocks, and can also be divided into three or less functional blocks or five or more functional blocks.

[0026] The data input unit 11 acquires input data input by the operator. The three-dimensional shape data of the object to be analyzed is input to the data input unit 11 as the object three-dimensional shape data. In addition, mesh data corresponding to the object three-dimensional shape data is input to the data input unit 11 as the object mesh data. The three-dimensional shape data is, for example, CAD data. In addition, the object mesh data is mesh data of a portion where a defect is not determined, and is mesh data to be analyzed. The object mesh data is produced by converting the object three-dimensional shape data.

[0027] In addition, the learning data of the machine learning model is input to the data input unit 11. For example, the learning mesh data is input to the data input unit 11. The learning mesh data is mesh data in which a defective part has been identified. For example, the learning mesh data is mesh data in which a defective part has been identified by a person through visual observation in the past. In addition, the three-dimensional shape data corresponding to the learning mesh data is input to the data input unit 11 as the learning three-dimensional shape data.

[0028] The learning mesh data includes the position of the defective part in the learning mesh data. For example, the learning mesh data includes the presence or absence of defects that need to be corrected in each mesh included in the learning mesh data. The presence or absence of defects is represented by two values ​​0 or 1. The mesh is composed of a plurality of nodes, so the position (coordinate) of the node constituting the mesh where the defect exists is determined as the defective part. In addition, the learning mesh data includes the part where the defective part exists in the learning mesh data. The part is classified into, for example, fillet, weld bead, no specific part, etc.

[0029] The feature quantity calculation unit 12 calculates the feature quantity of the mesh data input to the data input unit 11. When the target mesh data is input, the feature quantity calculation unit 12 calculates the feature quantity of the target mesh data based on the nodes and links constituting the target mesh data. For example, the feature quantity calculation unit 12 calculates the feature quantity of the mesh for each mesh included in the target mesh data using the coordinates of the nodes constituting the mesh and the connection relationship between the nodes. The feature quantity of the mesh is a value that quantitatively represents the feature of the mesh. In addition, when the learning mesh data is input, the feature quantity calculation unit 12 calculates the feature quantity of the learning mesh data based on the nodes and links constituting the learning mesh data. For example, the feature quantity calculation unit 12 calculates the feature quantity of the mesh for each mesh included in the learning mesh data.

[0030] The feature quantity of the mesh includes the feature quantity of the mesh surface and the feature quantity around the mesh. The feature quantity of the mesh surface includes, for example, the attribute value of the mesh surface and the geometric feature quantity with other mesh surfaces. The attribute value of the mesh surface includes the area of ​​the mesh surface, the angle of the mesh surface, and the length of each side of the mesh surface. The angle of the mesh surface is, for example, the maximum angle and the minimum angle of the internal angle of the mesh surface. In addition, the attribute value of the mesh surface includes a value representing the shape of the mesh surface. The shape of the mesh surface is either a triangle or a quadrilateral. The geometric feature quantity with other mesh surfaces includes the angle (dihedral angle) between the mesh surface and other mesh surfaces, and the distance between the mesh surface and other mesh surfaces. In addition, the feature quantity around the mesh includes the geometric feature quantity and statistical feature quantity around the mesh. The geometric feature quantity around the mesh includes the curvature of the mesh. For example, it is Gaussian curvature. In addition, the statistical feature quantity around the mesh is the density of the nodes constituting the mesh and the histogram of the normal relationship around the mesh.

[0031] In addition, the feature quantity calculation unit 12 calculates the feature quantity of the three-dimensional shape data input to the data input unit 11. The feature quantity of the three-dimensional shape data is a value that quantitatively represents the characteristics of the three-dimensional shape data. For example, the feature quantity of the three-dimensional shape data includes the geometric feature quantity of the surface and the edge contained in the three-dimensional shape data. The geometric feature quantity of the surface contained in the three-dimensional shape data includes the area, aspect ratio, volume, and curvature of the surface. In addition, the geometric feature quantity of the edge contained in the three-dimensional shape data includes the curvature and length of the edge. In addition, the feature quantity of the three-dimensional shape data includes the geometric feature quantity of the structure of the three-dimensional shape data. The geometric feature quantity of the structure of the three-dimensional shape data is, for example, the distance of the R-shaped bend contained in the three-dimensional shape data. The feature quantity calculation unit 12 calculates the feature quantity of the object three-dimensional shape data when the object three-dimensional shape data corresponding to the object mesh data is input. In addition, the feature quantity calculation unit 12 calculates the feature quantity of the learning three-dimensional shape data when the learning three-dimensional shape data corresponding to the learning mesh data is input.

[0032] In addition, the feature quantity calculation unit 12 calculates the feature quantity between the mesh data and the three-dimensional shape data corresponding to the mesh data input to the data input unit 11. The feature quantity between the mesh data and the three-dimensional shape data is, for example, the distance between the mesh included in the mesh data and the corresponding part in the three-dimensional shape data. When the object mesh data and the object three-dimensional shape data are input, the feature quantity calculation unit 12 calculates the feature quantity between the object mesh data and the object three-dimensional shape data. In addition, when the learning mesh data and the learning three-dimensional shape data are input, the feature quantity calculation unit 12 calculates the feature quantity between the learning mesh data and the learning three-dimensional shape data.

[0033] The learning unit 13 causes the machine learning model to learn using the training data to generate a learned model. The learned model is a model that has been learned in advance by machine learning so as to obtain appropriate output data for certain input data. The training data used in this embodiment is training data produced based on learning mesh data in which defective parts have been identified.

[0034] The learning unit 13 associates the learning mesh data input to the data input unit 11 with the feature quantity calculated by the feature quantity calculation unit 12 to generate training data in which the position of the defective part in the learning mesh data is associated with the feature quantity of the learning mesh data when the defective part is generated in the learning mesh data. For example, the learning unit 13 associates each node constituting the mesh included in the learning mesh data with the feature quantity of the mesh to generate training data. Thus, in the training data produced, the nodes constituting the mesh with defects in the learning mesh data (nodes with defects) are associated with the feature quantity of the mesh. The nodes (coordinates) with defects correspond to the positions of the defects. In addition, in the training data, the nodes constituting the mesh without defects in the learning mesh data (nodes without defects) are associated with the feature quantity of the mesh.

[0035] The learning unit 13 causes the machine learning model to learn using the generated training data. Specifically, the learning unit 13 causes the machine learning model to learn using the training data obtained by associating the position of the defective portion in the learning mesh data with the feature amount of the learning mesh data when the defective portion is generated in the learning mesh data, thereby generating a learned model for estimating the position of the defective portion.

[0036] The learning unit 13 first defines a function representing an input-output relationship to create a machine learning model. The machine learning model is a neural network with a hierarchical structure consisting of an input layer including one or more neurons, an output layer, and at least one intermediate layer. The feature quantity calculated by the feature quantity calculation unit 12 is input to the input layer as input data. For example, the feature quantity of each grid contained in the learning grid data is input. The output layer outputs whether there is a defective part for each node as output data corresponding to the input data. That is, whether it is the location where the defective part exists is output for each node. Whether there is a defective part is represented by two values ​​0 or 1.

[0037] The learning unit 13 uses the machine learning model and the training data to repeatedly change the function so that the sum of the squares of the errors between the output value (prediction value) of the machine learning model and the value (correct answer value) assigned by the training data is minimized. Specifically, the learning unit 13 learns in the following manner: the error between the prediction value output from the machine learning model based on the input data and the correct answer value of the training data is calculated for each node, and the coupling coefficient (weight) between each layer of the neural network is repeatedly updated to minimize the sum of the squares of the errors. When the learning unit 13 outputs the learned model obtained by learning, that is, the learned weight data, the learned model is stored in the database 2.

[0038] In addition, the type of neural network used by the learning unit 13 is not particularly limited, but in a conventional neural network, the number of combinations becomes huge, so a method called a graph convolutional neural network is used in this embodiment. For example, the nodes and connections constituting the grid included in the grid data are set as a graph structure, and in each intermediate layer of the neural network, the object node that becomes the object and the neighboring nodes adjacent to the object node are used to perform weight calculations.

[0039] In addition, in the present embodiment, it is assumed that the machine learning model uses training data based on the feature amount of the learning mesh data for learning, but it is not limited to this. In the present embodiment, the training data used for learning may also be training data obtained by associating the feature amount of the learning three-dimensional shape data and / or the feature amount between the learning mesh data and the learning three-dimensional shape data with the position of the defective part in the learning mesh data in addition to the feature amount of the learning mesh data. For example, the learning unit 13 associates the feature amount of the learning mesh data when a defective part is generated in the learning mesh data and the feature amount between the learning mesh data and the learning three-dimensional shape data corresponding to the learning mesh data with the position of the defective part in the learning mesh data to generate training data, and causes the machine learning model to learn using the generated training data.

[0040] In addition, in the present embodiment, the learned model for estimating the position of the defective part is generated using training data obtained by associating the position of the defective part in the learning mesh data with the feature amount of the learning mesh data, but the present invention is not limited thereto. In the present embodiment, the learned model for estimating the position of the defective part is generated using training data based on the position of the defective part in the learning mesh data. That is, the learning unit 13 associates the position of the defective part in the learning mesh data with the feature amount of the learning mesh data when the defective part is generated in the learning mesh data to generate the training data. Then, the learning unit 13 causes the machine learning model to learn using the generated training data.

[0041] In addition, the learned model for estimating the location where a defect exists can also be generated using training data obtained by associating feature quantities of the learning three-dimensional shape data and / or feature quantities between the learning mesh data and the learning three-dimensional shape data with the location where the defect exists in the learning mesh data in addition to the feature quantities of the learning mesh data.

[0042] The output unit 14 estimates the location or position of the defective part in the target mesh data using the learned model generated by the learning unit 13, and outputs the location or position of the defective part to the operator. The output unit 14 inputs the feature amount of the target mesh data as input data to the learned model for estimating the location of the defective part, and outputs the location of the defective part in the target mesh data from the learned model as output data corresponding to the input data. For example, when the feature amount of each mesh included in the target mesh data is input as input data, the learned model outputs the location of the defective part in the target mesh data. The location is, for example, any one of a fillet, a weld bead, and an unspecified location.

[0043] In addition, the output unit 14 inputs the feature amount of the object mesh data as input data to the learned model for estimating the position of the defective part, and outputs the position of the defective part in the object mesh data using the learned model. When the feature amount of the object mesh data is input as input data, the learned model outputs whether it is the position of the defective part for each node as output data. The output unit 14 detects the defective part by estimating the position of the defective part based on the output result of the learned model. Then, the output unit 14 outputs the coordinates of the node estimated as the position of the defective part to the operator. In the present embodiment, the output unit 14 draws the defective part in the three-dimensional model based on the object mesh data and displays it on the display. For example, the output unit 14 displays the node of the defective part in the three-dimensional model based on the object mesh data in a drawing method that is emphasized compared with the node without the defective part. The emphasized drawing method includes, for example, drawing with different colors and drawing with different graphics.

[0044] Here, use Figure 3 The following is an example of the display of the output results of the defective part. Figure 3 FIG. 2 is a diagram showing an example of a display of an output result of a defective part according to the present embodiment. Figure 3 In the example, a three-dimensional model based on the target mesh data is drawn and displayed by the nodes constituting the target mesh data. Moreover, nodes with defects, such as a node P with a defect in the target mesh data, are displayed in a drawing method different from nodes without defects.

[0045] In addition, in the present embodiment, the output unit 14 uses the feature amount of the object mesh data as input data to estimate the location or position of the defective part in the object mesh data using the learned model, but the present invention is not limited thereto. The output unit 14 may use the feature amount of the object three-dimensional shape data and / or the feature amount between the object mesh data and the object three-dimensional shape data as input data in addition to the feature amount of the object mesh data to estimate the location or position of the defective part in the object mesh data using the learned model.

[0046] In addition, in the present embodiment, the output unit 14 may first use the feature amount of the object mesh data as input data to estimate the candidate location of the defective part in the object mesh data using the learned model, thereby locking the candidate area of ​​the defective part. Then, after estimating the candidate location, the output unit 14 may also use the feature amount of the object mesh data at the candidate location as input data, the feature amount of the object three-dimensional shape data at the candidate location and / or the feature amount between the object mesh data at the candidate location and the object three-dimensional shape data as input data to estimate the location of the defective part in the object mesh data.

[0047] As described above, in this embodiment, a machine learning model is used to detect defective parts in target mesh data, thereby making it possible to detect defective parts more efficiently than an algorithm that detects defective parts by logic based on predetermined rules.

[0048] The database 2 stores various data. For example, the database 2 stores a learned model. In the present embodiment, the database 2 stores a learned model for estimating the position of a defective part and a learned model for estimating the location of a defective part. In addition, the database 2 stores input mesh data and three-dimensional shape data. In addition, the database 2 stores training data for learning the machine learning model.

[0049] Furthermore, in the present embodiment, the database 2 is a database installed in the defective portion output device 1 , but the present invention is not limited thereto, and the database 2 may be a database existing outside the defective portion output device 1 .

[0050] Next, use Figure 4 An example of a process for learning the machine learning model according to the present embodiment will be described. Figure 41 is a flowchart showing an example of a learning process of a machine learning model according to the present embodiment. In step S1, the controller 10 obtains input learning data. The learning data includes learning mesh data and learning three-dimensional shape data corresponding to the learning mesh data. The learning mesh data includes data on the location or position of a defective portion. In step S2, the controller 10 calculates a feature quantity. For example, the controller 10 calculates a feature quantity of the learning mesh data, a feature quantity of the learning three-dimensional shape data, and a feature quantity between the learning mesh data and the learning three-dimensional shape data.

[0051] In step S3, the controller 10 generates training data based on the learning data acquired in step S1 and the feature amount calculated in step S2. The generated training data is training data used in machine learning for estimating the location of a defective part and training data used in machine learning for estimating the position of a defective part. The training data used in machine learning for estimating the location of a defective part is training data obtained by associating the location of a defective part in the learning mesh data with the feature amount of the learning mesh data when a defective part is generated in the learning mesh data, the feature amount of the learning three-dimensional shape data, and the feature amount between the learning mesh data and the learning three-dimensional shape data. In addition, the training data used in machine learning for estimating the position of a defective part is training data obtained by associating the position of a defective part in the learning mesh data with the feature amount of the learning mesh data when a defective part is generated in the learning mesh data, the feature amount of the learning three-dimensional shape data, and the feature amount between the learning mesh data and the learning three-dimensional shape data.

[0052] In step S4, the controller 10 causes the machine learning model to learn using the training data generated in step S3. Through machine learning, the controller 10 generates a learned model that has been learned to estimate the location of the defective portion, and a learned model that has been learned to estimate the location of the defective portion. In step S5, the controller 10 outputs the learned model obtained by learning in step S4. Figure 4 In the embodiment, the feature quantity of the learning mesh data, the feature quantity of the learning three-dimensional shape data, and the feature quantity between the learning mesh data and the learning three-dimensional shape data are calculated and used for machine learning, but it is not limited to this. It is also possible to calculate only the feature quantity of the learning mesh data and use it for machine learning.

[0053] Next, use Figure 5 An example of the process of the defective portion output method involved in this embodiment is described. Figure 5FIG. 1 is a flowchart showing an example of the process of the defective portion output method according to the present embodiment. Figure 5 In step S11, the controller 10 obtains the input target mesh data. In step S12, the controller 10 calculates the feature amount of the target mesh data.

[0054] In step S13, the controller 10 inputs the feature quantity calculated in step S12 as input data to the learned model. Figure 4 The learned model obtained by learning in the object mesh data is, for example, a learned model for estimating the location of a defective part. In step S14, the controller 10 uses the learned model to estimate whether the defective part is located at each node included in the object mesh data. In step S15, the controller 10 outputs the estimation result estimated in step S14. For example, the controller 10 outputs the coordinates of the node estimated to be the location of the defective part to the operator. In addition, in the present embodiment, the location of the defective part can also be estimated using the learned model for estimating the location of the defective part, and the location of the defective part can be output to the operator.

[0055] Next, use Figure 6 An example of the process of the defective portion output method involved in this embodiment is described. Figure 6 FIG. 1 is a flowchart showing an example of the process of the defective portion output method according to the present embodiment. Figure 6 In step S21, the controller 10 obtains the input object mesh data and object three-dimensional shape data. In step S22, the controller 10 calculates the feature quantity based on the input data obtained in step S21. For example, the controller 10 calculates the feature quantity of the object mesh data, the feature quantity of the object three-dimensional shape data, and the feature quantity between the object mesh data and the object three-dimensional shape data.

[0056] In step S23, the controller 10 inputs the feature quantity calculated in step S22 as input data to the learned model. In step S24, the controller 10 uses the learned model to estimate whether the defective part is located at each node of the object mesh data. In step S25, the controller 10 outputs the estimation result estimated in step S24. For example, the controller 10 outputs the coordinates of the node estimated as the location where the defective part exists to the operator. In addition, in the present embodiment, the learned model for estimating the location where the defective part exists can also be used to estimate the location where the defective part exists, and the location where the defective part exists can be output to the operator.

[0057] As described above, in this embodiment, the controller calculates the feature amount of the object mesh data based on the nodes and links constituting the object mesh data to be processed by converting the three-dimensional shape data, inputs the feature amount of the object mesh data as input data to the learned model, and outputs the part or position of the defective part to be corrected in the object mesh data using the learned model, wherein the learned model is obtained by learning using training data obtained by associating the part or position of the defective part in the learning mesh data in which the defective part has been identified with the feature amount of the learning mesh data when the defective part is generated in the learning mesh data. Thus, the three-dimensional model for analysis can be corrected efficiently.

[0058] In addition, in the present embodiment, the controller calculates the feature quantity between the object mesh data and the object three-dimensional shape data based on the nodes and links constituting the object mesh data and the object three-dimensional shape data corresponding to the object mesh data, inputs the feature quantity between the object mesh data and the object three-dimensional shape data as input data to the learned model, and outputs the part or position of the defective part in the object mesh data that needs to be corrected, using the learned model, the learned model is learned using training data obtained by associating the part or position of the defective part in the learning mesh data with the feature quantity between the learning mesh data and the learning three-dimensional shape data corresponding to the learning mesh data when the defective part is generated in the learning mesh data. Thus, the three-dimensional model for analysis can be corrected more accurately and efficiently.

[0059] In addition, in the present embodiment, the controller inputs the feature amount of the object mesh data as input data into the learned model, uses the learned model to output a candidate portion of the object mesh data where a defective portion that needs to be corrected exists, inputs the feature amount of the object mesh data at the candidate portion and the feature amount between the object mesh data and the object three-dimensional shape data as input data into the learned model, and uses the learned model to output the position of the defective portion that needs to be corrected exists in the object mesh data. In this way, the three-dimensional model for analysis can be corrected more accurately and efficiently.

[0060] In addition, in the present embodiment, the feature values ​​of the object mesh data and the learning mesh data include the feature values ​​of the mesh surface of the mesh contained in the respective mesh data and the feature values ​​around the mesh. Thus, the defective part of the three-dimensional model for analysis can be detected using the feature values ​​of the mesh surface of the mesh contained in the three-dimensional model for analysis and the feature values ​​around the mesh.

[0061] In addition, in the present embodiment, the feature quantity between the object mesh data and the object three-dimensional shape data is the distance between the mesh contained in the object mesh data and the corresponding part in the object three-dimensional shape data. Thus, it is possible to detect the defective part of the three-dimensional model used for analysis based on the difference between the mesh data and the design shape data before mesh conversion.

[0062] In addition, the embodiments described above are embodiments for easy understanding of the present invention, and are not embodiments for limiting the present invention. Therefore, the elements disclosed in the above embodiments are intended to include all design changes and equivalents belonging to the protection scope of the present invention.

[0063] Description of Reference Numerals

[0064] 1: defective part output device; 10: controller; 11: data input unit; 12: feature quantity calculation unit; 13: learning unit; 14: output unit; 2: database.

Claims

1. A defective part output method, wherein a controller is used to output a defective part of mesh data generated by converting three-dimensional shape data, wherein: The controller performs the following processing: Calculating a feature amount of the target mesh data based on nodes and links constituting the target mesh data to be analyzed; Inputting the feature quantity of the object mesh data as input data into the learned model; Outputting the location or position of the defective part that needs to be corrected in the target mesh data using the learned model, The learned model is obtained by learning using training data, wherein the training data is obtained by establishing an association between the location or position of the defective part in the learning mesh data in which the defective part has been identified and the feature value of the learning mesh data when the defective part is generated in the learning mesh data.

2. The defective portion output method according to claim 1, wherein: The controller performs the following processing: calculating a feature amount between the object mesh data and the object three-dimensional shape data based on nodes and links constituting the object mesh data and object three-dimensional shape data corresponding to the object mesh data; inputting a feature quantity between the object mesh data and the object three-dimensional shape data as input data into the learned model; Outputting the location or position of the defective part that needs to be corrected in the target mesh data using the learned model, The learned model is obtained by learning using training data obtained by establishing an association between the location or position of the defective portion in the learning mesh data and the following feature quantity, which is a feature quantity between the learning mesh data and the learning three-dimensional shape data corresponding to the learning mesh data when a defective portion occurs in the learning mesh data.

3. The defective portion output method according to claim 2, wherein: The controller performs the following processing: Inputting the feature quantity of the object mesh data as input data into the learned model; Outputting a candidate location of a defective portion that needs to be corrected in the object mesh data using the learned model; inputting a feature quantity of the object mesh data at the candidate part and a feature quantity between the object mesh data and the object three-dimensional shape data as input data into the learned model; The learned model is used to output a position of a defective portion in the target mesh data that needs to be corrected.

4. The defective portion output method according to claim 1 or 2, wherein: The feature quantities of the target mesh data and the learning mesh data include feature quantities of mesh surfaces of meshes included in the respective mesh data and feature quantities of mesh peripheries.

5. The defective portion output method according to claim 2 or 3, wherein: The feature amount between the object mesh data and the object three-dimensional shape data is a distance between a mesh included in the object mesh data and a corresponding portion in the object three-dimensional shape data.

6. A defective portion output device, comprising a controller, for outputting a defective portion of mesh data generated by converting three-dimensional shape data, wherein: The controller performs the following processing: Calculating a feature amount of the target mesh data based on nodes and links constituting the target mesh data to be analyzed; Inputting the feature quantity of the object mesh data as input data into the learned model; Outputting the location or position of the defective part that needs to be corrected in the target mesh data using the learned model, The learned model is obtained by learning using training data, wherein the training data is obtained by establishing an association between the location or position of a defective part in learning mesh data in which a defective part has been identified and a feature value of the learning mesh data when the defective part is generated in the learning mesh data.

Images