Design system, computer mounting method, and computer program

Abstract

To suppress generation of an inappropriate shape such as a collapsed shape, when a structure such as a motor is automatically designed.SOLUTION: A design system includes a generator for generating a shape of a tangible structure, and an evaluator for evaluating the characteristics of the structure on the basis of the shape generated by the generator. The generator is configured to generate graph data in which the shape of the structure is represented by a graph having a node and an edge. The evaluator is configured to acquire the graph data from the generator, and to evaluate the characteristics of the structure, on the basis of the graph data.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to a design system, a computer-implemented method, and a computer program for a structure such as a motor. [Background technology]

[0002] Non-Patent Document 1 discloses an automated design system for interior permanent magnet synchronous motors. This automated design system includes a generative model that generates an image showing the shape of the rotor of the motor, and a predictive model that predicts motor characteristics from the generated image. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Yusei Shimizu, "Robust Optimization of IPMSM Efficiency Using Deep Generative Models", IEEJ Study Group Materials.SA, IEEJ Static Machine Study Group [ed.], 2022 (80-100), 7-12, 2022-09-30 [Non-Patent Document 2] Yusei Shimizu, "Technical Commentary by Motor Researchers", [online], [Retrieved August 7, 2023], Internet<URL:https: / / yuyumoyuyu.com / 2023 / 01 / 15 / paperintroduction2 / > Summary of the Invention

[0004] 10(A), the automated design system of Non-Patent Document 1 includes a generative model 310 that generates an image 400 showing the shape of a motor, and a characteristic prediction model 320 that predicts motor characteristics 450 from the generated image 400. This automated design system can automatically design an optimized structure (optimum solution) of a motor by repeatedly generating a shape and predicting (evaluating) the characteristics of the generated shape.

[0005] However, in the automatic design system of Non-Patent Document 1, the generative model may generate an unmanufacturable shape (a collapsed shape) as shown in FIG. 10(B), and this shape may be selected as the optimal solution.

[0006] Therefore, when automatically designing a structure such as a motor, it is desirable to prevent the generation of inappropriate shapes such as collapsed shapes.

[0007] Since images have too much freedom of expression, there is a risk that the generative model will generate shapes that cannot be manufactured. Therefore, by adopting a representation format with reduced freedom of expression, it is possible to prevent the generation of inappropriate shapes.

[0008] In this disclosure, a graph is used as an expression format with limited freedom of expression. The graph here is a graph in graph theory. A graph is composed of a set of nodes (vertices) and a set of edges (sides) connecting the nodes.

[0009] Although a graph has a lower degree of freedom of expression than an image, it has a sufficient degree of freedom of expression to express a two-dimensional or three-dimensional shape of a structure. Therefore, a graph is suitable as a form of expression for a generated shape, and can prevent an inappropriate shape such as a collapsed shape from being generated.

[0010] One aspect of the present disclosure is a system. The disclosed system may be a design system including a generator that generates a shape of a tangible structure, and an evaluator that evaluates characteristics of the structure based on the shape generated by the generator. The generator may be configured to generate graph data that represents the shape of the structure as a graph having nodes and edges. The evaluator may be configured to obtain the graph data from the generator and evaluate characteristics of the structure based on the graph data.

[0011] Another aspect of the present disclosure is a method. The disclosed method may be a computer-implemented method executed by a computer. The disclosed method may comprise generating a shape of a tangible structure and evaluating a characteristic of the structure based on the generated shape. Generating the shape of the structure may include generating graph data representing the shape of the structure as a graph having nodes and edges. Evaluating the characteristic of the structure may include evaluating the characteristic of the structure based on the graph data.

[0012] Another aspect of the disclosure is a computer program. The disclosed computer program may be a computer program that causes a computer to execute the steps of generating a shape of a tangible structure and evaluating a characteristic of the structure based on the generated shape. Generating the shape of the structure may include generating graph data that represents the shape of the structure as a graph having nodes and edges. Evaluating the characteristic of the structure may include evaluating the characteristic of the structure based on the graph data.

[0013] Further details will be described in the following embodiments. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a configuration diagram of the design system. [Diagram 2] FIG. 2 is a perspective view of the motor. [Diagram 3] FIG. 3 is a diagram showing the topology of one pole of the rotor. [Figure 4] FIG. 4 is a flowchart of the graph data generation process. [Diagram 5] FIG. 5 is an explanatory diagram of the graph data. [Figure 6] FIG. 6 is an explanatory diagram of the graph data. [Figure 7] FIG. 7 is an explanatory diagram of the graph data. [Figure 8] FIG. 8 is a diagram showing a variation of the generation of coordinates. [Figure 9] FIG. 9 shows a variation of the evaluator. [Figure 10] FIG. 10 is a diagram showing a design system that generates an image and an example of an image of a collapsed shape. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] <1. Overview of the design system, computer implementation method, and computer program>

[0016] (1) A design system according to an embodiment may include a generator that generates a shape of a tangible structure, and an evaluator that evaluates characteristics of the structure based on the shape generated by the generator. The generator may be configured to generate graph data that represents the shape of the structure as a graph having nodes and edges. The evaluator may be configured to acquire the graph data from the generator and evaluate characteristics of the structure based on the graph data. Although a graph has a lower degree of freedom of expression than an image, it has a sufficient degree of freedom of expression to represent the shape of a structure. Therefore, a graph is suitable as a form of expression for a shape generated by a generator, and can suppress the generation of an inappropriate shape such as a collapsed shape.

[0017] (2) It is preferable that the graph data has a matrix representing the graph and a feature vector indicating at least the position coordinates of the nodes included in the graph. In this case, the shape of the structure can be sufficiently represented.

[0018] (3) It is preferable that the feature vector further includes material data indicating a material associated with the node. In this case, the shape of the structure can be expressed including the material.

[0019] (4) The material associated with the node may include a permanent magnet, in which case the node may represent a structure including a permanent magnet.

[0020] (5) The material data may include data indicating the magnetization direction of the permanent magnet. In this case, the magnetization direction of the permanent magnet may also be expressed.

[0021] (6) The node may represent a vertex contained in the structure.

[0022] (7) The node may represent a partial area contained in the structure.

[0023] (8) The evaluator may include an evaluation model that has been machine-learned to output characteristics of the structure when the graph data indicating the shape of the structure is input.

[0024] (9) The evaluator may include a converter that converts the graph data representing the shape of the structure into an image representing the shape of the structure, and an evaluation model that has been machine-learned to output characteristics of the structure when the image representing the shape of the structure is input. The image output from the converter may be provided as an input to the evaluation model.

[0025] (10) The structure may be an electric machine.

[0026] (11) The structure may include a permanent magnet.

[0027] (12) The characteristics may include electromagnetic characteristics of the structure.

[0028] (13) The structure may be a motor, and the characteristics may include motor characteristics.

[0029] (14) A method according to an embodiment may be a computer-implemented method executed by a computer. The computer-implemented method may include generating a shape of a tangible structure, and evaluating a characteristic of the structure based on the generated shape. Generating the shape of the structure may include generating graph data that represents the shape of the structure as a graph having nodes and edges. Evaluating the characteristic of the structure may include evaluating the characteristic of the structure based on the graph data.

[0030] (15) A computer program according to an embodiment may cause a computer to generate a shape of a tangible structure and evaluate a characteristic of the structure based on the generated shape. Generating the shape of the structure may include generating graph data that represents the shape of the structure as a graph having nodes and edges. Evaluating the characteristic of the structure may include evaluating the characteristic of the structure based on the graph data.

[0031] <2. Examples of design systems, computer implementation methods, and computer programs>

[0032] Hereinafter, examples according to the embodiment will be described in detail with reference to the drawings. FIG. 1 shows a design system 100 according to the embodiment. The design system 100 designs the shape of a structure. The shape to be designed is, for example, a physical two-dimensional or three-dimensional shape of a tangible structure. The shape may be the shape of a part of the structure, or the shape of the entire structure.

[0033] The structure may be, for example, an electric machine. Electric machines here are a general term for rotating electric machines (rotating machines) such as motors and generators, and stationary electric machines (stationary machines) such as transformers, rectifiers, and switching control devices. Electric machines are machines whose electromagnetic characteristics can be evaluated in the design process.

[0034] The structure may, by way of example, comprise a magnet, for example an electric machine such as a motor.

[0035] The design system 100 according to the embodiment includes a generator 110 and an evaluator 120. The generator 110 generates a shape of a structure. The evaluator 120 evaluates characteristics of the structure based on the shape generated by the generator 110.

[0036] The design system 100 may be configured by one or more computers having a processor and a memory. The processor may be a CPU or a GPU. The memory is connected to the processor. The memory includes, for example, a primary storage device and a secondary storage device. The primary storage device is, for example, a RAM. The secondary storage device is, for example, a hard disk drive (HDD) or a solid-state drive (SSD). The memory includes a computer program executed by the processor. The processor reads and executes the computer program stored in the memory. The computer program has a program code for causing the computer to operate as the design system 100. That is, the computer program has a program code for causing the computer to operate as the generator 110 and the evaluator 120.

[0037] The design system 100 can output an optimized shape by repeatedly generating a shape and evaluating it. For example, the design system 100 uses the determiner 150 to determine whether the characteristics evaluated (predicted) by the evaluator 120 satisfy a predetermined requirement, and if the requirements are satisfied, outputs a shape having the characteristics as an optimized structure. If the characteristics evaluated by the evaluator 120 do not satisfy the predetermined requirement, the design system 100 regenerates and evaluates the shape. The design system 100 automatically outputs an optimized shape by repeatedly generating various shapes and evaluating them.

[0038] The properties evaluated by the evaluator 120 are properties of the structure having the shape generated by the generator 110. The properties may be, for example, rotor properties or stationary properties. The properties may be, for example, electromagnetic properties. If the structure is a motor, the properties may be motor properties. The motor properties may include, for example, any one or a combination of motor speed-torque characteristics, motor efficiency, torque ripple, and iron losses.

[0039] In the following description, the design system 100 according to the embodiment is used to design a motor, as an example. Motors are used in various products such as electric vehicles, drones, aircraft, and industrial robots. The motor to be designed by the design system 100 is, as an example, an interior permanent magnet synchronous motor (IPMSM).

[0040] As shown in Fig. 2, interior permanent magnet synchronous motor 10 is a type of motor that uses permanent magnets 15. Interior permanent magnet synchronous motor 10 includes stator 11 and rotor 12 provided inside stator 11. Permanent magnets 15 are embedded in rotor 12. Interior permanent magnet synchronous motor 10 shown in Fig. 2 is, as an example, an 8-pole, 48-slot distributed winding IPMSM.

[0041] Stator 11 is made of electromagnetic steel and has multiple slots 16 formed in the circumferential direction. Copper wire (not shown) is wound in slots 16. Air exists in the slots 16. Therefore, there are three types of materials to be considered in designing stator 11: electromagnetic steel, copper, and air.

[0042] Rotor 12 is configured by embedding permanent magnets 15 in a rotor body made of electromagnetic steel plate. In rotor 12, a region where air exists (flux barrier) is formed in a position adjacent to permanent magnet 15. Therefore, there are three types of materials to be considered in designing rotor 12: electromagnetic steel plate, permanent magnet, and air.

[0043] Here, the design subject is rotor 12, and the structure of stator 11 (stator structure) is assumed to be determined in advance. The external shape of rotor 12 is also assumed to be determined in advance. Therefore, the design subject here is the shape and arrangement, etc., of the permanent magnets and air in rotor 12. Note that the external shapes of stator 11 and rotor 12 may also be designed.

[0044] Furthermore, since each pole of the rotor 12 has the same structure, when designing the rotor 12, it is sufficient to design only one pole of the rotor 12.

[0045] 3(A) to (C) show examples of two-dimensional topology for one pole of the rotor 12. The positions, numbers, shapes, etc. of the permanent magnets and flux barriers (air) differ in each topology shown in FIG.

[0046] In the design system 100 according to the embodiment, the generator 110 generates and outputs graph data 200 indicating the two-dimensional shape of one pole of the rotor 12 shown in Fig. 3. The evaluator 120 acquires the graph data 200 output from the generator 110, and evaluates the motor characteristics of a motor (IPMSM) equipped with the rotor 12 having the generated shape based on the graph data 200.

[0047] 4 shows an example of the procedure of the process S400 for generating the shape of a structure, which is executed by the generator 110. As described above, the generator 110 according to the embodiment generates the shape of the rotor 12 as graph data 200 expressed by a graph. The graph includes nodes (vertices) and edges (sides). Although the graph has a lower degree of freedom of expression than an image, it has a sufficient degree of freedom of expression to express a two-dimensional or three-dimensional shape of a structure. Therefore, the graph is suitable as a form of expression for the shape to be generated, and can prevent the generation of an inappropriate shape such as a collapsed shape.

[0048] As shown in FIGS. 5(C) and 5(D), the graph data 200 can be represented by a matrix representing the graph (see FIG. 5(C)) and a feature vector (see FIG. 5(D)).

[0049] A matrix that represents a graph is, for example, an adjacency matrix or a connectivity matrix. An adjacency matrix indicates the adjacency relationship between nodes in a graph due to edges (whether nodes are adjacent to each other via edges). An adjacency matrix is ​​a form of graph expression that focuses on the adjacency relationship between nodes. On the other hand, a connectivity matrix is ​​a form of graph expression that focuses on the connectivity between nodes and edges. A connectivity matrix indicates the connectivity between nodes and edges (whether a node and an edge are connected).

[0050] Here, as an example, an adjacency matrix (see FIG. 5(C)) is used as a matrix representing a graph. The adjacency matrix in FIG. 5(C) shows the adjacency relationship of nodes "1," "2," "3," "4," etc. in the graph shown in FIG. 5(B). FIG. 5(B) represents the shape of rotor 12 shown in FIG. 5(A) as a graph having nodes and edges.

[0051] The adjacency matrix shown in Figure 5(C) indicates that node "1" and node "2" are adjacent, node "2" and node "4" are adjacent, node "4" and node "3" are adjacent, and node "3" and node "1" are adjacent.

[0052] As shown in Fig. 5(D), the feature vector indicates at least the position coordinates of the nodes included in the graph. The position coordinates indicate the position where the node exists in the shape of the rotor 12 (structure). In the feature vector shown in Fig. 5(D), the coordinates of node "1" are represented by (x1, y1), the coordinates of node "2" are represented by (x2, y2), the coordinates of node "3" are represented by (x3, y3), and the coordinates of node "4" are represented by (x4, y4).

[0053] The feature vector shown in Fig. 5(D) has material data indicating the material associated with each node. That is, the feature vector shown in Fig. 5(D) is data in which the position coordinates and material data are associated with the node.

[0054] The material associated with a node is, for example, the material constituting the region adjacent to the node in the rotor 12 (structure). The material data indicates what the material of the region adjacent to the node is. In the feature vector of FIG. 5(D), the material data indicates whether the material of the region adjacent to the node is air, a permanent magnet, or both. In FIG. 5(D), the material data is described by a one-hot vector. Note that the material of the rotor 12 also includes electromagnetic steel sheets, but since the node is always adjacent to the electromagnetic steel sheets, they are omitted in the material data shown in FIG. 5(D). By omitting the electromagnetic steel sheets, the data size can be reduced. Note that the electromagnetic steel sheets may be explicitly indicated in the material data.

[0055] In the feature vector of FIG. 5(D), the material data of node "1" is represented by (a,d)=(1,0), indicating that node "1" is adjacent to the region of air a, but not adjacent to the region of permanent magnet d. The material data of node "2" is represented by (a,d)=(1,1), indicating that node "2" is adjacent to the region of air a and permanent magnet d. The material data of node "3" is represented by (a,d)=(1,0), indicating that node "3" is adjacent to the region of air a, but not adjacent to the region of permanent magnet d. The material data of node "4" is represented by (a,d)=(1,1), indicating that node "4" is adjacent to the region of air a and permanent magnet d.

[0056] FIG. 6 shows another example of the feature vector. In the feature vector of FIG. 6, the material data is data d mx Includes. mxis a variable representing the magnetization direction, and the direction ranging from -90 degrees to 90 degrees is indicated by a value ranging from 0 to 1. x is an identifier for each permanent magnet included in the rotor 12 with one pole, and takes on values ​​from 1 to X (X is the number of permanent magnets included in the rotor 12 with one pole). When the rotor 12 with one pole has three permanent magnets (x=1, x=2, x=3), the material data is (a, d m1 ,d m2 ,d m3 )

[0057] In the feature vector of Figure 6, the material data of node "1" is (a, d m1 ,d m2 ,d m3 ) = (1,0,0,0), which indicates that node "1" is adjacent to the region of air a and is not adjacent to any of the regions of permanent magnets (x = 1 to 3). The material data of node "2" is (a,d m1 ,d m2 ,d m3 )=(1,d m1 ,0,0), node "2" is adjacent to the region of air a and the permanent magnet at x=1, and the magnetization direction of the permanent magnet at x=1 is d m1 The material data for node "3" is (a,d m1 ,d m2 ,d m3 ) = (1,0,0,0), which indicates that node "3" is adjacent to the region of air a and is not adjacent to any of the regions of permanent magnets (x = 1 to 3). The material data for node "4" is (a,d m1 ,d m2 ,d m3 )=(1,d m1 ,0,0), node "4" is adjacent to the region of air a and the permanent magnet at x=1, and the magnetization direction of the permanent magnet at x=1 is d m1 This indicates that.

[0058] 5 and 6, the nodes of the graph indicate vertices included in the rotor 12 (structure). Here, a vertex refers to a point where two or more lines representing a shape intersect, or a point where they are connected in a non-linear manner. In this case, the shape of the rotor 12 (structure) is represented by its vertices as the nodes of the graph, and the straight lines between the vertices as the edges of the graph. In other words, the graphs shown in FIGS. 5 and 6 represent the structure with a focus on the vertices.

[0059] FIG. 7 shows an example of a graph expressing a structure with a focus on a partial region included in the structure. Here, a partial region is a two-dimensional or three-dimensional partial region constituting the structure. A partial region is a small region obtained by dividing a structure into a plurality of regions, like an element in the finite element method. The shape of the partial region may be one type, or may be multiple types. A partial region may be set for each material of the structure.

[0060] In FIG. 7, the nodes of the graph indicate partial regions included in the rotor 12 (structure). Also, in FIG. 7, the nodes of the graph indicate partial regions indicating the regions of the permanent magnets included in the rotor 12 and partial regions indicating the regions of air present in the rotor 12. Also, the edges of the graph indicate that the partial regions indicated by the nodes are adjacent to each other. In the graph representation shown in FIG. 7, the representation of the electromagnetic steel sheet regions, which represent most of the region of the rotor 12, is omitted, and the regions of the permanent magnets and air are represented by nodes and edges. By omitting the representation of the electromagnetic steel sheet regions, the data size can be reduced.

[0061] In addition, one permanent magnet area may be represented by multiple nodes (multiple partial areas representing permanent magnets), and one air area (flux barrier) may be represented by multiple nodes (multiple partial areas representing air).

[0062] In the case of a graph representation focusing on a partial region as in Fig. 7, the feature vector may include data indicating either or both of the shape and size of the partial region, in addition to the position coordinates and material data (which may include the magnetization direction) of the nodes shown in Figs. 5 and 6. Therefore, in the case of the graph representation shown in Fig. 7, the feature vector is more likely to be high-dimensional data than in the graph representations of Figs. 5 and 6, but the number of edges can be reduced. Also, in the case of the graph representation in Fig. 7, it is easy to add new edges and new shapes can be easily represented, making it suitable for generating new shapes (shapes of new topologies).

[0063] On the other hand, in the case of the graph representations in Figures 5 and 6, the number of edges is large, but the feature vector can be made low-dimensional. Also, in the case of the graph representations in Figures 5 and 6, it is difficult to add new edges, so they are suitable for generating typical shapes (typical topology shapes as shown in Figures 3(A) to (C)).

[0064] Returning to Fig. 4, the generation process S400 by the generator 110 will be described. First, the generator 110 selects a generation topology (step S401). A plurality of selectable generation topologies are set in advance in the generator 110. The plurality of selectable generation topologies are, for example, the three rotor topologies shown in Figs. 3(A) to (C).

[0065] The generation topology may be selected by accepting a generation topology selection operation performed by a user operation, or the generator 110 may automatically determine a generation topology from among a plurality of generation topologies based on an appropriate rule or randomly.

[0066] Each of the multiple pre-set generating topologies is set with an adjacency matrix (a matrix representing a graph) and values ​​other than the node position coordinates in the feature vector (e.g., material data associated with each node). Therefore, when a generating topology is selected in step S401, values ​​other than the node position coordinates in the adjacency matrix and the feature vector are determined according to the selected generating topology (step S402). In other words, by selecting a generating topology, values ​​other than the node position coordinates are determined in the graph data 200 indicating the shape of the rotor 12 (structure). In this way, the generating topology indicates graph data in which only the node position coordinates are variables (undetermined values). Note that the number of nodes in each generating topology may be a fixed value.

[0067] The generator 110 generates node position coordinates in the selected generated topology (step S403). By generating the position coordinates of each node, graph data 200 (generated data) consisting of an adjacency matrix and feature vectors is generated. An example of a method for generating node position coordinates will be described later.

[0068] The generator 110 verifies the generated graph data (generated data) (step S404). This verification is performed based on a predetermined verification criterion to determine whether the generated graph data (generated data) properly represents the shape of the rotor 12. The verification criterion is, for example, a criterion indicating a case where the shape is inappropriate. The criterion indicating a case where the shape is inappropriate is, for example, "each edge intersects," "the node position coordinates are outside the specified area," and "the distance between the coordinates of two nodes is equal to or less than a predetermined value."

[0069] Edges (straight lines that indicate the shape of a structure) should not cross, so if any edges cross, the shape is improper.

[0070] Since the node position coordinates should be within a preset designated area in each generated topology, if they are outside the designated area, the shape is inappropriate. The designated area may be set within the rotor 12.

[0071] From the viewpoint of strength and manufacturability, it is preferable that the width of the electromagnetic steel sheet is equal to or greater than a certain value, but if the two nodes are too close, the width of the electromagnetic steel sheet becomes too narrow. Therefore, if the distance between the coordinates of the two nodes is equal to or less than a certain value, the shape is inappropriate.

[0072] If it is determined as a result of the verification in step S404 that the generated shape is inappropriate (problematic) (step S405), the process returns to step S403, and node position coordinates in the selected generated topology are regenerated.

[0073] If it is determined as a result of the verification in step S404 that the generated shape is appropriate (no problem) (step S405), the generator 110 outputs the generated graph data 200 (generated data). Therefore, only appropriate graph data 200 is output from the generator 110. The output graph data 200 is provided to the evaluator 120.

[0074] Fig. 8 shows variations in the generation of coordinates in step S403. Fig. 8(A) shows an example in which the coordinates (x, y) of a node are generated as random numbers based on an appropriate random number seed. Fig. 8(B) shows an example in which the coordinates are sampled using a machine learning model such as kernel density estimation or a Gaussian mixture model. Fig. 8(C) shows an example of generation using a deep generative model such as variational autoencoding. Note that data other than the coordinates in the feature vector may also be generated together with the coordinates.

[0075] Fig. 9 shows a variation of the evaluator 120 shown in Fig. 1. The characteristic evaluation of the structure based on the graph data 200 may be either of Fig. 9(A) and Fig. 9(B).

[0076] 9(A) includes a first characteristic evaluation model 121 (trained model) that has been machine-learned to output the characteristics of a motor equipped with a rotor 12 having a shape shown in the graph data 200 when the graph data 200 is input. The first characteristic evaluation model 121 has been machine-learned (deep learning) using the graph data 200 indicating the shape of the rotor 12 and characteristic data of the motor equipped with the rotor 12 as learning data.

[0077] 9(A), graph data 200 output from the generator 110 is provided to the input of a first characteristic evaluation model 121. When the graph data 200 is input, the first characteristic evaluation model 121 can output motor characteristics.

[0078] The evaluator 120 shown in FIG. 9(B) includes a converter 125 and a second characteristic evaluation model 126. The converter 125 converts the graph data 200 into an image. The image represents the shape of the rotor 12 indicated by the graph data 200 in a predetermined image data format (e.g., bitmap). The second characteristic evaluation model 126 is machine-learned to output the characteristics of a motor including the rotor 12 having the shape indicated by the image when the image indicating the shape of the rotor 12 is input. The second characteristic evaluation model 126 is machine-learned (deep learning) using the image indicating the shape of the rotor 12 and characteristic data of the motor including the rotor 12 as learning data.

[0079] 9(B), the graph data 200 output from the generator 110 is provided to the converter 125. The converter 125 converts the provided graph data 200 into an image and outputs it. The image output from the converter 125 is provided to the input of a second characteristic evaluation model 126. When the image is input, the second characteristic evaluation model 126 can output the motor characteristics.

[0080] In the case of the evaluator 120 in FIG. 9(A), the converter 125 of the evaluator 120 in FIG. 9(B) is not necessary, so the configuration can be simplified. Also, in the case of the evaluator 120 in FIG. 9(B), an image-based evaluation model 126 can be used. In characteristic evaluation using a machine learning model, image input is common, so there are many excellent existing techniques for characteristic evaluation. Therefore, by adopting an evaluation model 126 for image input as in FIG. 9(B), it is possible to utilize many existing techniques for characteristic evaluation.

[0081] As the motor characteristic evaluation model 126 with image input shown in Fig. 9(B), for example, a prediction model (deep learning model) described in Non-Patent Documents 1 and 2 can be used. As the motor characteristic evaluation model 121 with graph input shown in Fig. 9(A), one constructed by machine learning (deep learning) using the input of the prediction model described in Non-Patent Documents 1 and 2 as graph data 200 instead of an image can be used.

[0082] It should be noted that data other than the graph data 200 or images (for example, operating conditions of the motor, other conditions of the motor) may also be input to the evaluation models 121, 126 in Figs. 9(A) and 9(B).

[0083] The present invention is not limited to the above-described embodiment, and various modifications are possible. [Explanation of symbols]

[0084] 10: Interior permanent magnet synchronous motor 11: Stator 12: Rotor 15: Permanent magnet 16: Slots 100: Design Systems 110: Generator 120: Evaluator 121: Motor characteristic evaluation model 125: Converter 126: Motor characteristic evaluation model 150: Judgment device 200: Graph data 310: Generative Models 320: Property prediction model 400:Image 450: Motor characteristics

Claims

1. A generator for generating a shape of a tangible structure; an evaluator for evaluating characteristics of the structure based on the shape generated by the generator; Equipped with The generator is configured to generate graph data that represents a shape of the structure as a graph having nodes and edges; The evaluator is configured to obtain the graph data from the generator and evaluate the property of the structure based on the graph data. Design system.

2. The graph data includes a matrix representing the graph and a feature vector indicating at least the position coordinates of the nodes included in the graph. The design system of claim 1 .

3. The feature vector further comprises material data indicative of a material associated with the node. The design system according to claim 2 .

4. The material associated with the node includes a permanent magnet. The design system according to claim 3 .

5. The material data includes data indicating a magnetization direction of the permanent magnet. The design system according to claim 4.

6. The nodes indicate vertices included in the structure. The design system of claim 1 .

7. The node indicates a partial area included in the structure. The design system of claim 1 .

8. The evaluator includes an evaluation model that is machine-learned to output the characteristics of the structure when the graph data indicating the shape of the structure is input. The design system of claim 1 .

9. The evaluator is a converter for converting the graph data representing the shape of the structure into an image representing the shape of the structure; an evaluation model trained by machine learning to output the characteristics of the structure when the image showing the shape of the structure is input; Equipped with The image output from the converter is provided to the input of the evaluation model. The design system of claim 1 .

10. The structure is an electric machine. The design system of claim 1 .

11. The structure includes a permanent magnet. The design system of claim 1 .

12. The properties include electromagnetic properties of the structure. The design system according to claim 10 or 11.

13. the structure is a motor, The characteristics include motor characteristics. The design system of claim 1 .

14. 1. A computer-implemented method performed by a computer, comprising: Generate the shape of a tangible structure; Evaluating properties of the structure based on the generated shape Prepare for this. generating a shape of the structure includes generating graph data that represents the shape of the structure as a graph having nodes and edges; Evaluating the characteristic of the structure includes evaluating the characteristic of the structure based on the graph data. Computer-implemented method.

15. On the computer, Generate the shape of a tangible structure; Evaluating properties of the structure based on the generated shape A computer program for causing a computer to execute the steps of: generating a shape of the structure includes generating graph data that represents the shape of the structure as a graph having nodes and edges; Evaluating the property of the structure includes evaluating the property of the structure based on the graph data. Computer program.

Images