Defective location output method and defective location output device

Abstract

A controller (10): calculates a feature amount of target mesh data on the basis of nodes and links which constitute the target mesh data serving as a target for an analysis process created by converting three-dimensional shape data; inputs, as input data, the feature amount of the target mesh data into a trained model; and uses the trained model to output a site or position where a defective location required to be corrected is present in the target mesh data. The trained model is trained by using teaching data which associates the site or position of a defective location in the training mesh data in which the defective location has already been specified, and the feature amount of the training mesh data when the defective location occurs in the training mesh data.

Description

Defective part output method and defective part output device

[0001] The present invention relates to a defect location output method and a defect location output device.

[0002] A technology is known in which a three-dimensional model based on three-dimensional CAD data is divided into a grid to generate a three-dimensional model for analysis, and the differences 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 are calculated and displayed to an operator (Patent Document 1).

[0003] Japanese Patent Application Laid-Open No. 2004-94663

[0004] However, the technology of Patent Document 1 does not display the areas that need to be corrected in the 3D model for analysis, so the operator must visually check the 3D model for analysis and identify the areas that need to be corrected, which results in the problem that the 3D model for analysis cannot be corrected efficiently.

[0005] The problem to be solved by the present invention is to provide a defect location output method and device that can efficiently correct a three-dimensional model for analysis.

[0006] The present invention solves the above problem by calculating the features of the target mesh data based on the nodes and links that make up the target mesh data that is the subject of analysis processing and that is created by converting three-dimensional shape data, inputting the features of the target mesh data as input data to a trained model, and using the trained model outputting the parts or positions of defective parts in the target mesh data that need to be corrected, and the trained model is trained using training data that associates the parts or positions of defective parts in training mesh data in which defective parts have already been identified with the features of the training mesh data when a defective part occurs in the training mesh data.

[0007] According to the present invention, a three-dimensional model for analysis can be efficiently corrected.

[0008] FIG. 1 is a block diagram showing a schematic configuration of a defect location output device according to an embodiment of the present invention; FIG. 2 is a diagram showing an example in which the defect location output device according to the present embodiment is applied; FIG. 3 is a diagram showing an example of an output result of a defect location according to the present embodiment; FIG. 4 is a flowchart showing an example of the procedure of a learning method for a machine learning model according to the present embodiment; FIG. 5 is a flowchart showing an example of the procedure of a defect location output method according to the present embodiment;

[0009] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0010] FIG. 1 is a block diagram showing a schematic configuration of a defect location output device according to an embodiment of the present invention. The defect location output device 1 is applied to a system that creates mesh data used in CAE (Computer Aided Engineering) analysis. In CAE analysis, structural mechanics, fluid dynamics, and other analyses are performed on mesh data created by converting design shape data of an object to be analyzed. The design shape data of the object to be analyzed is three-dimensional shape data created using software such as CAD. Mesh data is data created by dividing the three-dimensional shape data into a grid. In other words, mesh data is data that represents the design shape of the object to be analyzed using a mesh. The mesh data includes data on the nodes and links that constitute the mesh. Furthermore, the nodes include data on the node number and coordinates for each node. The node coordinates indicate the position of the node in three-dimensional space. The link data is data on the connection relationship between the nodes. In each mesh, multiple nodes are connected to form a triangle or a rectangle.

[0011] Mesh data is converted from three-dimensional shape data of an object to be analyzed using existing automatic meshing software functions or the like. Conventionally, mesh data created using existing automatic meshing software functions or the like has defects that require correction. Therefore, a person had to visually check the three-dimensional model based on the mesh data displayed on a display, identify the defects, and correct them, which made corrections inefficient. In this embodiment, the defect location output device 1 automatically detects and outputs defects in the mesh data, thereby enabling efficient correction of the mesh data.

[0012] An example of application of the defect location device according to this embodiment will be described with reference to FIG. 2. This is a diagram showing an example of application of the defect location output device according to this embodiment. FIG. 2( a) shows an example of a three-dimensional model based on three-dimensional shape data to be analyzed by CAE. 3D shape data such as that shown in FIG. 2( a) is created using software such as CAD. FIG. 2( b) shows an example of a three-dimensional model based on mesh data converted from the three-dimensional shape data. Mesh data such as that shown in FIG. 2( b) is created using an automatic meshing software function or the like. FIG. 2( c) shows defect locations P1 and P2 occurring in the three-dimensional model based on the mesh data shown in FIG. 2( b). The defect location output device 1 automatically detects defect locations such as defect locations P1 and P2 and outputs them to an operator.

[0013] Defects in mesh data often occur when analyzing structures such as automobiles, which are made up of multiple assembled parts. Because these structures are assembled, there are many fillets and beads, which leads to many defects in the mesh data. Examples of defects in mesh data include defects where a part that would be a ridge in the three-dimensional shape data is not a ridge, defects where the shape of a part that would be a fillet or bead in the three-dimensional shape data is distorted, and defects at merging points.

[0014] The defect location output device 1 includes a controller 10 and a database 2. The controller 10 includes a computer having hardware and software, and this computer includes a ROM storing a program, a CPU that executes the program stored in the ROM, and a RAM that functions as an accessible storage device. Note that, as the operating circuit, an MPU, DSP, ASIC, FPGA, etc. can be used instead of or in addition to the CPU.

[0015] The controller 10 detects defective parts by estimating the location or position of defective parts in the mesh data using the trained model, and outputs the detected defective parts. The controller 10 includes functional blocks, such as a data input unit 11, a feature calculation unit 12, a learning unit 13, and an output unit 14, and executes each function through cooperation between software and hardware for realizing each function or executing each process. Note that in this embodiment, the functions of the controller 10 are divided into four blocks, and the functions of each functional block are described, but the functions of the controller 10 do not necessarily have to be divided into four blocks, and may be divided into three or fewer functional blocks, or five or more functional blocks.

[0016] The data input unit 11 acquires input data input by an operator. Three-dimensional shape data of an object to be analyzed is input to the data input unit 11 as target three-dimensional shape data. Mesh data corresponding to the target three-dimensional shape data is also input to the data input unit 11 as target mesh data. The three-dimensional shape data is, for example, CAD data. The target mesh data is mesh data in which no defect location has been identified and is the mesh data that is the target of analysis processing. The target mesh data is created by conversion from the target three-dimensional shape data.

[0017] Furthermore, training data for the machine learning model is input to the data input unit 11. For example, training mesh data is input to the data input unit 11. The training mesh data is mesh data in which defect locations have already been identified. For example, the training mesh data is mesh data in which defect locations have been visually identified by a human in the past. Furthermore, three-dimensional shape data corresponding to the training mesh data is input to the data input unit 11 as training three-dimensional shape data.

[0018] The training mesh data includes the locations of defective parts in the training mesh data. For example, the training mesh data includes, for each mesh included in the training mesh data, whether or not there is a defect in the mesh that requires correction. The presence or absence of a defect is expressed as a binary value of 0 or 1. Because a mesh is made up of multiple nodes, the position (coordinates) of a node that makes up a defective mesh is identified as the defective part. The training mesh data also includes the part in the training mesh data where the defective part is located. The part is classified, for example, into fillet, bead, no specific part, etc.

[0019] The feature calculation unit 12 calculates feature quantities of mesh data input to the data input unit 11. When target mesh data is input, the feature calculation unit 12 calculates feature quantities of the target mesh data based on the nodes and links that make up the target mesh data. For example, the feature calculation unit 12 calculates mesh feature quantities for each mesh included in the target mesh data using the coordinates of the nodes that make up the mesh and the connections between the nodes. Mesh feature quantities are values ​​that quantitatively indicate mesh features. Furthermore, when training mesh data is input, the feature calculation unit 12 calculates feature quantities of the training mesh data based on the nodes and links that make up the training mesh data. For example, the feature calculation unit 12 calculates mesh feature quantities for each mesh included in the training mesh data.

[0020] The mesh features include the features of the mesh faces and the features of the mesh surroundings. The features of the mesh faces include, for example, the attribute values ​​of the mesh faces and the geometric features relative to other mesh faces. The attribute values ​​of the mesh faces include the area of ​​the mesh faces, the angle of the mesh faces, and the length of each side of the mesh faces. The angle of the mesh faces is, for example, the maximum and minimum interior angles of the mesh faces. The attribute values ​​of the mesh faces also include a value indicating the shape of the mesh faces. The shape of the mesh faces is either a triangle or a quadrangle. The geometric features relative to other mesh faces include the angle (dihedral angle) between the mesh face and other mesh faces and the distance between the mesh face and other mesh faces. The features of the mesh surroundings also include the geometric features and statistical features of the mesh surroundings. The geometric features of the mesh surroundings include the curvature of the mesh, for example, Gaussian curvature. The statistical features of the mesh surroundings are the density of the nodes constituting the mesh and a histogram of the normal relationship around the mesh.

[0021] The feature calculation unit 12 also calculates feature quantities of the three-dimensional shape data input to the data input unit 11. The feature quantities of the three-dimensional shape data are values ​​that quantitatively indicate the features of the three-dimensional shape data. For example, the feature quantities of the three-dimensional shape data include geometric feature quantities of faces and ridges included in the three-dimensional shape data. The geometric feature quantities of faces included in the three-dimensional shape data include the area, aspect ratio, volume, and curvature of the faces. The geometric feature quantities of ridges included in the three-dimensional shape data include the curvature and length of the ridges. The feature quantities of the three-dimensional shape data also include geometric feature quantities of the structure of the three-dimensional shape data. An example of the geometric feature quantity of the structure of the three-dimensional shape data is the radius of curvature included in the three-dimensional shape data. When target three-dimensional shape data corresponding to the target mesh data is input, the feature calculation unit 12 calculates the feature quantities of the target three-dimensional shape data. When training three-dimensional shape data corresponding to the training mesh data is input, the feature calculation unit 12 calculates the feature quantities of the training three-dimensional shape data.

[0022] Furthermore, the feature amount calculation unit 12 calculates feature amounts between the mesh data and the three-dimensional shape data corresponding to the mesh data input to the data input unit 11. The feature amount between the mesh data and the three-dimensional shape data is, for example, the distance between a mesh included in the mesh data and a corresponding location in the three-dimensional shape data. When target mesh data and target three-dimensional shape data are input, the feature amount calculation unit 12 calculates feature amounts between the target mesh data and the target three-dimensional shape data. When training mesh data and training three-dimensional shape data are input, the feature amount calculation unit 12 calculates feature amounts between the training mesh data and the training three-dimensional shape data.

[0023] The learning unit 13 trains a machine learning model using training data to generate a trained model. A trained model is a model that has been trained in advance by machine learning so that appropriate output data can be obtained for certain input data. The training data used in this embodiment is training data created based on training mesh data in which defect locations have already been identified.

[0024] The learning unit 13 associates the training mesh data input to the data input unit 11 with the feature values ​​calculated by the feature value calculation unit 12, and generates training data that associates the positions of defective areas in the training mesh data with the feature values ​​of the training mesh data when a defective area occurs in the training mesh data. For example, the learning unit 13 associates each node constituting a mesh included in the training mesh data with the feature values ​​of the mesh to generate training data. As a result, in the generated training data, nodes constituting a mesh with a defect in the training mesh data (nodes with a defective area) are associated with the feature values ​​of the mesh. The node (coordinates) with the defective area corresponds to the position of the defective area. Furthermore, the training data associates nodes constituting a mesh without a defect in the training mesh data (nodes without a defective area) with the feature values ​​of the mesh.

[0025] The learning unit 13 trains a machine learning model using the generated training data. Specifically, the learning unit 13 trains the machine learning model using training data that associates the positions of defective parts in the training mesh data with feature amounts of the training mesh data when a defective part occurs in the training mesh data, and generates a trained model for estimating the positions of defective parts.

[0026] The learning unit 13 first defines a function that indicates the 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, an output layer, and at least one intermediate layer, each consisting of one or more neurons. The features calculated by the feature calculation unit 12 are input to the input layer as input data. For example, the features of each mesh included in the training mesh data are input. The output layer outputs, as output data corresponding to the input data, whether or not there is a defect for each node. In other words, it outputs, for each node, whether or not there is a defect location. Whether or not there is a defect is expressed as a binary value, 0 or 1.

[0027] The learning unit 13 uses the machine learning model and training data to repeatedly change the function so as to minimize the sum of squares of the error between the output value (predicted value) of the machine learning model and the value (correct value) given by the training data. Specifically, the learning unit 13 calculates, for each node, the error between the predicted value output from the machine learning model based on the input data and the correct value of the training data, and performs learning by repeatedly updating the coupling coefficients (weights) between each layer of the neural network so as to minimize the sum of squares of the error. When the learning unit 13 outputs the trained model, i.e., the trained weight data, it stores the trained model in the database 2.

[0028] Furthermore, the type of neural network used by the learning unit 13 is not particularly limited, but since the number of combinations becomes enormous in a normal neural network, this embodiment uses a technique called a graph convolutional neural network. For example, the nodes and links constituting the meshes contained in the mesh data are treated as a graph structure, and in each intermediate layer of the neural network, weights are calculated using the target node and the neighboring nodes in the vicinity of the target node.

[0029] Furthermore, in the present embodiment, the machine learning model is trained using training data based on the feature quantities of the training mesh data, but this is not limited to this. In the present embodiment, the training data used for training may be training data in which, in addition to the feature quantities of the training mesh data, feature quantities of the training 3D shape data and / or feature quantities between the training mesh data and the training 3D shape data are associated with the location of a defect in the training mesh data. For example, when a defect occurs in the training mesh data, the training unit 13 generates training data by associating the feature quantities of the training mesh data and feature quantities between the training mesh data and the training 3D shape data corresponding to the training mesh data with the location of the defect in the training mesh data, and then trains the machine learning model using the generated training data.

[0030] Furthermore, in this embodiment, a trained model for estimating the location of a defect is generated using training data that associates the location of a defect in the training mesh data with the feature values ​​of the training mesh data, but this is not limited to this. In this embodiment, a trained model for estimating the location of a defect is generated using training data based on the location of a defect in the training mesh data. That is, the learning unit 13 generates training data by associating the location of a defect in the training mesh data with the feature values ​​of the training mesh data when a defect occurs in the training mesh data. Then, the learning unit 13 trains a machine learning model using the generated training data.

[0031] Furthermore, the trained model for estimating the location of a defect may be generated using training data that associates, in addition to the features of the training mesh data, the features of the training three-dimensional shape data and / or the features between the training mesh data and the training three-dimensional shape data with the location of a defect in the training mesh data.

[0032] The output unit 14 uses the trained model generated by the learning unit 13 to estimate the part or position of the defective part in the target mesh data and outputs the part or position of the defective part to an operator. The output unit 14 inputs the feature amounts of the target mesh data as input data to the trained model for estimating the part of the defective part, and outputs the part of the defective part in the target mesh data from the trained model as output data corresponding to the input data. For example, when the feature amounts for each mesh included in the target mesh data are input as input data, the trained model outputs the part of the defective part in the target mesh data. The part is, for example, a fillet, a bead, or no specific part.

[0033] The output unit 14 also inputs the feature quantities of the target mesh data as input data into a trained model for estimating the location of a defect, and uses the trained model to output the location of the defect in the target mesh data. When the feature quantities of the target mesh data are input as input data, the trained model outputs, for each node, output data indicating whether or not a defect exists. The output unit 14 detects the defect by estimating the location of the defect based on the output result of the trained model. The output unit 14 then outputs the coordinates of the node estimated to be the location of the defect to the operator. In this embodiment, the output unit 14 renders the defect in a three-dimensional model based on the target mesh data and displays it on a display. For example, the output unit 14 displays nodes in the three-dimensional model based on the target mesh data that have a defect in a more emphasized manner than nodes that do not have a defect. The emphasized rendering manner includes, for example, rendering in a different color or a different shape.

[0034] Here, an example of the display of the output result of the defect part will be described with reference to Fig. 3. Fig. 3 is a diagram showing an example of the display of the output result of the defect part according to this embodiment. In Fig. 3, a 3D model based on the target mesh data is displayed by being drawn using the nodes that make up the target mesh data. Then, among the target mesh data, a node that has a defect part, such as node P that has a defect part, is displayed in a drawing format that is different from that of a node that does not have a defect part.

[0035] In the present embodiment, the output unit 14 uses the feature amounts of the target mesh data as input data and the trained model to estimate the portion or location of the defective part in the target mesh data, but this is not limited to this. The output unit 14 may also use the feature amounts of the target 3D shape data and / or the feature amounts between the target mesh data and the target 3D shape data as input data in addition to the feature amounts of the target mesh data and use the trained model to estimate the portion or location of the defective part in the target mesh data.

[0036] In the present embodiment, the output unit 14 may first use the feature amounts of the target mesh data as input data and use the trained model to estimate candidate regions in the target mesh data where the defect may be located, thereby narrowing down the candidate regions where the defect may be located. After estimating the candidate regions, the output unit 14 may use the feature amounts of the target mesh data in the candidate regions as input data, as well as the feature amounts of the target 3D shape data and / or the feature amounts between the target mesh data and the target 3D shape data.

[0037] As described above, in this embodiment, a machine learning model is used to detect defects in target mesh data, which allows for more efficient detection of defects than algorithms that use logic based on predetermined rules.

[0038] The database 2 stores various data. For example, the database 2 stores trained models. In this embodiment, the database 2 stores trained models trained to estimate the location of a defect, and trained models trained to estimate the location of a defect. The database 2 also stores input mesh data and three-dimensional shape data. The database 2 also stores training data for training the machine learning model.

[0039] In this embodiment, the database 2 is a database installed in the defect location output device 1, but is not limited to this and may be a database located outside the defect location output device 1.

[0040] Next, an example of a procedure for training a machine learning model according to this embodiment will be described with reference to FIG. 4 . FIG. 4 is a flowchart illustrating an example of a procedure for training a machine learning model according to this embodiment. In step S1, the controller 10 acquires input training data. The training data includes training mesh data and training three-dimensional shape data corresponding to the training mesh data. The training mesh data includes data on a portion or position where a defective portion is present. In step S2, the controller 10 calculates feature amounts. For example, the controller 10 calculates feature amounts of the training mesh data, feature amounts of the training three-dimensional shape data, and feature amounts between the training mesh data and the training three-dimensional shape data.

[0041] In step S3, the controller 10 generates training data based on the training data acquired in step S1 and the feature values ​​calculated in step S2. The generated training data includes training data used in machine learning to estimate a location of a defect and training data used in machine learning to estimate a position of the defect. The training data used in machine learning to estimate a location of a defect is training data that associates a location of the defect in the training mesh data with a feature value of the training mesh data, a feature value of the training three-dimensional shape data, and a feature value between the training mesh data and the training three-dimensional shape data when the defect occurs in the training mesh data. The training data used in machine learning to estimate a position of a defect is training data that associates a location of the defect in the training mesh data with a feature value of the training mesh data, a feature value of the training three-dimensional shape data, and a feature value between the training mesh data and the training three-dimensional shape data when the defect occurs in the training mesh data.

[0042] In step S4, the controller 10 trains a machine learning model using the training data generated in step S3. Through machine learning, the controller 10 generates a trained model trained to estimate the location of a defective portion and a trained model trained to estimate the portion of a defective portion. In step S5, the controller 10 outputs the trained model trained in step S4. Note that in FIG. 4 , the feature amounts of the training mesh data, the feature amounts of the training 3D shape data, and the feature amounts between the training mesh data and the training 3D shape data are calculated and used for machine learning, but this is not limiting, and only the feature amounts of the training mesh data may be calculated and used for machine learning.

[0043] Next, an example of the procedure of the defect location output method according to this embodiment will be described with reference to Fig. 5. Fig. 5 is a flowchart showing an example of the procedure of the defect location output method according to this embodiment. In Fig. 5, the controller 10 outputs the defect location using mesh data. In step S11, the controller 10 acquires input target mesh data. In step S12, the controller 10 calculates the feature amount of the target mesh data.

[0044] In step S13, the controller 10 inputs the feature amount calculated in step S12 as input data to the trained model. The trained model is the trained model trained in FIG. 4 , for example, a trained model for estimating the location of a defect. In step S14, the controller 10 uses the trained model to estimate whether or not a defect exists for each node included in the target 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 defect to an operator. Note that in this embodiment, the location of the defect may be estimated using a trained model for estimating the location of the defect, and the location of the defect may be output to an operator.

[0045] Next, an example of the procedure of the defect location output method according to this embodiment will be described with reference to FIG. 6 . FIG. 6 is a flowchart illustrating an example of the procedure of the defect location output method according to this embodiment. In FIG. 6 , the controller 10 outputs a defect location using mesh data and three-dimensional shape data. In step S21, the controller 10 acquires input target mesh data and target three-dimensional shape data. In step S22, the controller 10 calculates feature amounts based on the input data acquired in step S21. For example, the controller 10 calculates feature amounts of the target mesh data, feature amounts of the target three-dimensional shape data, and feature amounts between the target mesh data and the target three-dimensional shape data.

[0046] In step S23, the controller 10 inputs the feature amount calculated in step S22 as input data to the trained model. In step S24, the controller 10 uses the trained model to estimate whether or not a defect exists for each node of the target 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 to be the location of the defect to the operator. Note that in this embodiment, the location of the defect may be estimated using a trained model for estimating the location of the defect, and the location of the defect may be output to the operator.

[0047] As described above, in this embodiment, the controller calculates feature quantities of the target mesh data based on the nodes and links that make up the target mesh data that is the subject of analysis processing and that was created by converting the three-dimensional shape data, inputs the feature quantities of the target mesh data as input data to the trained model, and uses the trained model to output parts or positions of defect parts in the target mesh data that need to be corrected, and the trained model is trained using training data that associates parts or positions of defect parts in the training mesh data, in which defect parts have already been identified, with the feature quantities of the training mesh data when a defect part occurs in the training mesh data. This allows the three-dimensional model for analysis to be corrected efficiently.

[0048] In this embodiment, the controller calculates feature quantities between the target mesh data and the target 3D shape data based on the nodes and links that make up the target mesh data and the target 3D shape data that corresponds to the target mesh data, inputs the feature quantities between the target mesh data and the target 3D shape data as input data to the trained model, and uses the trained model to output the parts or positions of defect areas in the target mesh data that need to be corrected, the trained model being trained using training data that associates the parts or positions of defect areas in the training mesh data with feature quantities between the training mesh data and the training 3D shape data that corresponds to the training mesh data when a defect occurs in the training mesh data. This enables the 3D model for analysis to be corrected more accurately and efficiently.

[0049] In this embodiment, the controller inputs the feature amounts of the target mesh data to the trained model as input data, uses the trained model to output candidate parts in the target mesh data that have defect parts that need to be corrected, inputs the feature amounts of the target mesh data and feature amounts between the target mesh data and the target three-dimensional shape data for the candidate parts as input data to the trained model, and uses the trained model to output the positions of the defect parts in the target mesh data that need to be corrected. This makes it possible to correct the three-dimensional model for analysis more accurately and efficiently.

[0050] In this embodiment, the feature amounts of the target mesh data and the training mesh data include feature amounts of the mesh faces of the meshes included in each mesh data and feature amounts of the mesh surroundings, thereby making it possible to detect defects in the 3D model for analysis using the feature amounts of the mesh faces of the meshes included in the 3D model for analysis and the mesh surroundings.

[0051] In this embodiment, the feature amount between the target mesh data and the target 3D shape data is the distance between the mesh included in the target mesh data and the corresponding location in the target 3D shape data, which allows detection of defects in the 3D model for analysis based on the difference between the mesh data and the design shape data before mesh conversion.

[0052] It should be noted that the above-described embodiments have been described to facilitate understanding of the present invention, and are not intended to limit the present invention. Therefore, each element disclosed in the above-described embodiments is intended to include all design modifications and equivalents that fall within the technical scope of the present invention.

[0053] REFERENCE SIGNS LIST 1... Defect location output device 10... Controller 11... Data input section 12... Feature amount calculation section 13... Learning section 14... Output section 2... Database

Claims

1. A defect location output method for outputting defect locations of mesh data generated by converting three-dimensional shape data by a controller, comprising the steps of: The controller: Calculating a feature amount of the target three-dimensional shape data and / or a feature amount between the target mesh data and the target three-dimensional shape data based on target mesh data to be analyzed and / or target three-dimensional shape data corresponding to the target mesh data; Inputting the feature amount of the target 3D shape data and / or the feature amount between the target mesh data and the target 3D shape data into a trained model as input data; Using the trained model, outputting a portion or position of a defective portion that needs to be corrected in the target mesh data; The trained model is A portion or position where a defect exists in the learning mesh data in which a defect portion has already been identified; and A defect output method in which, when a defect occurs in the training mesh data, training data is performed using teacher data that associates features of training three-dimensional shape data corresponding to the training mesh data and / or features between the training mesh data and the training three-dimensional shape data.

2. The controller: calculating a feature amount of the target mesh data, a feature amount of the target three-dimensional shape data, and / or a feature amount between the target mesh data and the target three-dimensional shape data based on the nodes and links constituting the target mesh data and the target three-dimensional shape data; Inputting, as input data, a feature of the target mesh data, a feature of the target three-dimensional shape data, and / or a feature between the target mesh data and the target three-dimensional shape data into the trained model; Using the trained model, outputting a portion or position of a defective portion that needs to be corrected in the target mesh data; The trained model is 2. The defect output method according to claim 1, wherein learning is performed using teacher data in which the features of the training mesh data, the features of the training three-dimensional shape data, and / or the features between the training mesh data and the training three-dimensional shape data when a defect occurs in the training mesh data are associated with a location or position in the training mesh data where a defect exists.

3. The controller: Calculating a feature amount of the target 3D shape data and a feature amount between the target mesh data and the target 3D shape data based on the target mesh data and the target 3D shape data; Inputting a feature of the target 3D shape data and a feature between the target mesh data and the target 3D shape data as input data into the trained model; Using the trained model, outputting a portion or position of a defective portion that needs to be corrected in the target mesh data; The trained model is 2. The defect output method according to claim 1, wherein learning is performed using teacher data that associates the features of the training three-dimensional shape data and the features between the training mesh data and the training three-dimensional shape data when a defect occurs in the training mesh data with the location or position of the defect in the training mesh data.

4. The controller: Inputting the feature amount of the target mesh data into the trained model as input data; Using the trained model, outputting candidate parts having defect parts that need to be corrected in the target mesh data; Inputting a feature amount of the target mesh data and a feature amount between the target mesh data and the target three-dimensional shape data in the candidate portion as input data into the trained model; The defect output method according to claim 2 , further comprising the step of outputting a location of a defect in the target mesh data that requires correction, using the trained model.

5. 5. The defect location output method according to claim 2, wherein the feature amounts of the target mesh data and the learning mesh data include feature amounts of mesh faces of meshes included in the respective mesh data and feature amounts of mesh surroundings.

6. A defect location output method according to any one of claims 1 to 5, wherein the feature between the target mesh data and the target three-dimensional shape data is the distance between a mesh contained in the target mesh data and a corresponding location in the target three-dimensional shape data.

7. A defect location output device including a controller for outputting defect locations of mesh data generated by converting three-dimensional shape data, The controller: Calculating a feature amount of the target three-dimensional shape data and / or a feature amount between the target mesh data and the target three-dimensional shape data based on target mesh data to be analyzed and / or target three-dimensional shape data corresponding to the target mesh data; Inputting the feature amount of the target 3D shape data and / or the feature amount between the target mesh data and the target 3D shape data into a trained model as input data; Using the trained model, outputting a portion or position of a defective portion that needs to be corrected in the target mesh data; The trained model is A portion or position where a defect exists in the learning mesh data in which a defect portion has already been identified; and A defect location output device that learns using teacher data that associates features of training three-dimensional shape data corresponding to the training mesh data when a defect occurs in the training mesh data, and / or features between the training mesh data and the training three-dimensional shape data.