Design system, command generation system, method, and computer program

Abstract

The present disclosure makes it possible to receive a language-based instruction from a user and execute shape modification (design change) of an object being designed in accordance with the language-based instruction. A design system according to the present disclosure comprises a processor that executes processing including: receiving, from a user, a language-based instruction relating to shape modification of object data representing the shape of an object being designed; generating, from the language-based instruction, a modification command for a shape modification process for modifying the object data related to the shape; and executing the shape modification process on the object data on the basis of the modification command.

Description

Design system, command generation system, method, and computer program

[0001] This disclosure relates to a design system, a command generation system, a method, and a computer program. This application claims priority to Japanese Application No. 2024-26816, filed February 26, 2024, and incorporates by reference all of the contents of said Japanese application.

[0002] In the design of a design object such as a motor, the designer generally conducts design development through trial and error.

[0003] JP 6-131184 A Japanese Patent No. 7341581 A Japanese Patent No. 7353696 A

[0004] 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

[0005] Deep learning technology is rapidly improving in the field of artificial intelligence. Large language models (LLMs) capable of natural language interaction are trained using large amounts of text data and can take specific actions based on vague human text instructions. However, because the training data for LLMs does not include detailed design data for specific design targets, such as motors, it is still difficult to directly utilize them for quantitative design strategy analysis.

[0006] Therefore, there is a need for a technology that can accept a user's linguistic instruction and modify the shape (change the design) of an object in accordance with the linguistic instruction.

[0007] One aspect of the present disclosure is a design system including a processor that performs operations including: receiving, from a user, a linguistic instruction regarding shape modification of target data indicating a shape of the target, generating, from the linguistic instruction, a modification command for a shape modification process that modifies the shape of the target data, and executing the shape modification process of the target data based on the modification command.

[0008] Another aspect of the present disclosure is a command generation system including a processor that performs operations including receiving, from a user, a linguistic instruction regarding shape modification of object data indicating a shape of the object, and generating, from the linguistic instruction, a modification command for a shape modification process that modifies the shape of the object data.

[0009] Another aspect of the present disclosure is a method, the method being a computer-implemented method executed by a computer, including receiving, from a user, a linguistic instruction regarding shape modification of target data indicating a shape of the target, generating, from the linguistic instruction, a modification command for a shape modification process that modifies the shape of the target data, and performing the shape modification process of the target data based on the modification command.

[0010] Another aspect of the present disclosure is a computer program that causes a computer to perform operations including: receiving, from a user, a linguistic instruction regarding shape modification of target data indicating a shape of the target, generating, from the linguistic instruction, a modification command for a shape modification process that modifies the shape of the target data, and executing the shape modification process of the target data based on the modification command.

[0011] Further details will be described in the following embodiments.

[0012] FIG. 1 is a configuration diagram of a design system. FIG. 2(A) is a cross-sectional view of one pole of an interior permanent magnet synchronous motor, and FIG. 2(B) is a diagram showing variations in rotor topology of an interior permanent magnet synchronous motor. FIG. 3 is a flowchart showing the processing procedure of a shape modification system. FIG. 4(A) shows a list of command names for modification commands, and FIG. 4(B) shows examples of modification commands. FIG. 5(A) shows a processing judgment LLM, FIG. 5(B) shows a JSON creation LLM, and FIG. 5(C) shows an interactive LLM. FIG. 6 is an explanatory diagram for setting the role of an LLM. FIG. 7 is a flowchart showing the processing procedure of the design system. FIG. 8 is a diagram showing an example of user input and an example of LLM output. FIG. 9 is a diagram showing an example of user input and an example of LLM output. FIG. 10 is a diagram showing an example of dialogue between the design system and a user. FIG. 11 is a configuration diagram of a modified example of the design system.

[0013] 1. Overview of design system, command generation system, method, and computer program

[0014] (1) A system according to an embodiment is a design system including a processor that executes operations including: receiving, from a user, a linguistic instruction regarding shape modification of target data indicating a shape of the target; generating, from the linguistic instruction, a modification command for a shape modification process that modifies the shape of the target data; and executing, based on the modification command, the shape modification process of the target data. With this design system, the modification command for the shape modification process is generated from the user's linguistic instruction, and therefore the shape modification process can be executed based on the modification command.

[0015] (2) Generating the correction command may include inputting the linguistic instruction to an artificial intelligence and obtaining, from the artificial intelligence, a correction command for shape correction corresponding to the linguistic instruction. The shape correction process may include correcting the shape of the target data based on the correction command by a shape correction system to which the correction command obtained from the artificial intelligence is input. The artificial intelligence may learn a plurality of correction commands preset for shape correction in the shape correction system.

[0016] (3) The artificial intelligence may select and output a plurality of correction commands for shape correction in accordance with the linguistic instruction from the plurality of correction commands that it has learned.

[0017] (4) The operation may further include providing the target data before modification to the artificial intelligence, and the artificial intelligence may select and output a plurality of modification commands from the plurality of learned modification commands for shaping the target data before modification in accordance with the linguistic instruction.

[0018] (5) The object may be a motor, and the modification command may be a command for modifying the shape of the object data indicating the shape of the motor.

[0019] (6) The modification command may be a command conforming to an input format for a shape modification process. The shape processing system may be configured to execute the shape modification process by inputting the modification command conforming to the input format. The artificial intelligence may have learned the input format of the modification command and may generate a modification command for modifying the shape in response to the linguistic instruction conforming to the input format.

[0020] (7) The operation performed by the processor may further include determining whether there is a correction command corresponding to the linguistic instruction, and if it is determined that there is no correction command corresponding to the linguistic instruction, prompting the user to re-input the linguistic instruction. Generating the correction command may be performed if it is determined that there is a correction command corresponding to the linguistic instruction.

[0021] (8) Determining whether there is a correction command corresponding to the linguistic instruction may include determining, by an artificial intelligence that has learned a plurality of correction commands for modifying a shape, whether there is a correction command corresponding to the linguistic instruction among the plurality of correction commands learned by the artificial intelligence.

[0022] (9) The operations performed by the processor may further include determining whether information required to generate the correction command is available in the linguistic instructions, and prompting the user to input the information if it is determined that the information is not available. The execution of the shape correction process may be performed if it is determined that the information is available.

[0023] (10) Determining whether the information necessary for generating the modification command is present in the language instructions may include determining, by artificial intelligence that has learned the format of modification commands for shape modification, whether the information necessary for generating the modification command according to the format is present in the language instructions.

[0024] (11) The modification command according to the format may include parameters for modifying a shape, and determining whether information required to generate the modification command according to the format is available in the language instructions may include determining whether information required to determine the parameters is available in the language instructions.

[0025] (12) The operation may further include determining, by an artificial intelligence that has learned a plurality of correction commands for shape correction, whether a correction command corresponding to the linguistic instruction is included in the plurality of correction commands learned by the artificial intelligence, and prompting the user to re-input the linguistic instruction if it is determined that a correction command corresponding to the linguistic instruction is not included in the plurality of correction commands learned by the artificial intelligence. Generating the correction command may be performed if it is determined that a correction command corresponding to the linguistic instruction is included in the plurality of correction commands learned by the artificial intelligence. Generating the correction command may include generating, by an artificial intelligence that has learned a format of correction commands for shape correction, a correction command corresponding to the linguistic instruction in accordance with the format. The artificial intelligence that has learned the format of correction commands for shape correction may determine whether the linguistic instruction contains all the information necessary to generate the correction command in accordance with the format. The operation may further include prompting the user to input the information if it is determined that the information is not included.

[0026] (13) The modification command according to the format may include parameters for modifying a shape, and determining whether information required to generate the modification command according to the format is available in the language instructions may include determining whether information required to determine the parameters is available in the language instructions.

[0027] (14) Generating the correction command from the linguistic instruction may be performed by a large-scale language model.

[0028] In any one of the above (1) to (14), the target may be a motor.

[0029] (15) A system according to an embodiment may be a command generation system that includes a processor that performs operations including receiving, from a user, a linguistic instruction regarding shape modification of target data indicating a shape of the target, and generating, from the linguistic instruction, a modification command for a shape modification process that modifies the shape of the target data.

[0030] (16) A method according to an embodiment may be a computer-implemented method executed by a computer. The computer-implemented method may include receiving, from a user, a linguistic instruction regarding shape modification of target data indicating a shape of the target, generating, from the linguistic instruction, a modification command for a shape modification process that modifies the shape of the target data, and executing the shape modification process of the target data based on the modification command.

[0031] (17) A computer program according to an embodiment causes a computer to perform operations including: receiving, from a user, a linguistic instruction regarding shape modification of target data indicating a shape of the target, generating, from the linguistic instruction, a modification command for a shape modification process that modifies the shape of the target data, and executing the shape modification process of the target data based on the modification command. The computer program can be stored in a computer-readable non-transitory storage medium.

[0032] 2. Examples of design system, command generation system, method, and computer program

[0033] Hereinafter, embodiments of the present invention will be described in more detail with reference to the drawings.

[0034] FIG. 1 shows a design system 100 according to an embodiment. The design system 100 outputs output data 40 in response to a user instruction 10. The design system 100 is used, for example, to design 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 portion of the structure, or the shape of the entire structure.

[0035] The design system 100 shown in FIG. 1 includes a command generation system 200 and a shape modification system 300. The command generation system 200 generates a modification command 20 to be given to the shape modification system 300 from an instruction 10 from a user such as a designer. The instruction 10 is an instruction from a user such as a designer. The instruction 10 may include an instruction regarding the modification of the target shape. The instruction may be a linguistic instruction in natural language. Note that the linguistic instruction is also called a "text instruction."

[0036] The shape modification system 300 executes a shape modification process to modify the shape of the object data 30 based on the modification command 20 generated by the command generation system 200. Here, "shape" refers to the geometric properties of an object, and shape modification refers to changing the geometric properties. Geometric properties refer to the shape, size, position, and other spatial properties of an object. Shape modification may be a change to the internal shape of the object, or a change to the external shape of the object. For example, shape modification may be a change to the internal shape of the object while maintaining the external shape of the object. Shape modification may also be called a design change or a structural change.

[0037] The target data 30 is digital data representing the shape of the target. The target data 30 is, for example, image data or CAD data representing the shape of the target. Upon receiving the modification command 20, the shape modification system 300 executes a shape modification process to modify the target data 30 in accordance with the modification command 20, and outputs the modified target data as output data 40.

[0038] 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 program code including instructions for causing a computer to operate as the design system 100. That is, the computer program has program code for causing a computer to operate as the command generation system 200 and the shape modification system 300.

[0039] In the following description, the object of design is, as an example, a motor. Motors are used in various products such as electric vehicles, drones, aircraft, and industrial robots. In the following description, the motor to be designed by the design system 100 is, as an example, an interior permanent magnet synchronous motor (IPMSM). Note that the object of design by the design system 100 is not limited to a motor, and may be other objects.

[0040] An IPMSM is a type of motor that uses permanent magnets. Figure 2(A) shows a cross section of one pole of an 8-pole, 48-slot distributed winding IPMSM. Because each pole of an IPMSM has a common structure, it is sufficient to design one pole when designing an IPMSM. As shown in Figure 2(A), the IPMSM comprises a stator 11 and a rotor 12 provided inside the stator 11. A permanent magnet 15 is embedded in the rotor 12.

[0041] The stator 11 is made of electromagnetic steel and has a plurality of slots 16 formed in the circumferential direction. Copper wire (not shown) is wound in the slots 16. Air exists in the slots 16.

[0042] The rotor 12 is configured by embedding permanent magnets 15 in a rotor body (rotor core) made of electromagnetic steel plate. In the rotor 12, a region where air exists (flux barrier) is formed in a position adjacent to the permanent magnets 15, etc.

[0043] Here, as an example, the design target is the rotor 12, and the structure of the stator 11 (stator structure) is assumed to be predetermined. Also, the outer shape of the rotor 12 is assumed to be predetermined.

[0044] Fig. 2(B) shows four examples (V, 2D, Nabla, 2U) of topologies of two-dimensional shapes for one pole of the rotor 12. Each of the topologies shown in Fig. 3(A) to (C) differs in the internal shape of the rotor 12, such as the shape and arrangement of the permanent magnets.

[0045] As an example, the shape modification system 300 uses the four topologies shown in FIG. 2B as target data 30 (reference shapes). Which of the multiple topologies is to be used as the target of shape modification may be specified by the user, or may be selected appropriately by the design system 100. Furthermore, in addition to the topologies shown in FIG. 2B, target data input by the user to the design system 100 may also be the target of shape modification. Furthermore, for re-modification, output data 40 output by the shape modification system 300 may also be the target of shape modification.

[0046] 3 shows the procedure for modifying a shape by the shape modification system 300. The shape modification system 300 receives input of a modification command 20 indicating how to perform the modification and target data 30 (input shape). As an example, the shape modification system 300 executes pre-processing S31, shape modification processing S32, and post-processing S33.

[0047] The preprocessing S31 extracts graph information from the input target data 30. Here, the target data 30 input to the shape modification system 300 is, for example, CAD data. The preprocessing S31 extracts graph information from the CAD data 30. The graph information is data that represents the shape of the object to be modified, such as the rotor 12, using a graph. The graph includes nodes (vertices) and edges (sides). As an example, a node of the graph indicates a vertex included in the rotor 12 (object). Here, a vertex refers to a point where two or more lines representing the shape intersect or are connected in a non-linear manner. In this case, the shape of the rotor 12 is represented by the vertex coordinates represented by the graph nodes, and the straight lines between the vertices represented by the graph edges. In other words, the shape of the rotor 12 is treated as a graph structure focusing on the coordinates of each vertex. Representing the shape as graph information facilitates shape modification.

[0048] The shape modification process S32 modifies the vertex coordinates of the graph structure in accordance with the modification command 20. This modifies the shape of the rotor 12, which is treated as a graph structure. In the shape modification process S32, the shape modification is performed by a rule-based algorithm for shape modification. The modification command 20 is, for example, a modification instruction for the rule-based algorithm. The rule-based algorithm is executed in accordance with the modification command 20. The type, content, and data format of the modification command 20 that can be accepted by this algorithm are set in advance. In other words, the shape modification system 300 needs to input a predetermined modification command 20 that complies with a predetermined data format (input format).

[0049] In post-processing S33, the graph information corrected in step S32 is decoded, restored to CAD information, and output as output data 40 (output shape).

[0050] FIG. 4A shows an example of a modification command 20. FIG. 4A shows the command names of six modification commands. The six command names include "move," "move_single_point," "rotate," "designate," "enlarge," and "approximate." The modification command 20 includes not only the name of the command to be executed, but also information such as parameters required to execute each shape modification process. The modification command 20 is expressed in JSON format, as an example. The JSON format is a type of text data format. The JSON format can include multiple members expressed as ("key":value). Each member includes a pair of key and value. The key indicates the name of the value, and the value indicates the value of the key.

[0051] The "move" command moves a specified part in a particular direction. The direction of movement by the "move" command can be specified in various formats, such as along the axis of symmetry, radially, angularly along the rotor, or any direction specified numerically. The direction of movement can be specified by parameters to the "move" command.

[0052] Fig. 4B shows an example of a modification command 20 in JSON format that includes a "move" command. In the modification command 20 in Fig. 4B, "command":"move" specifies a modification action, and in this case, indicates that the modification action is move. The specification of the modification action also specifies the command name ("move").

[0053] "Topology": "Nabla" specifies the rotor topology to be modified, and in this example, indicates that the topology to be modified is "Nabla" (see FIG. 2B). "Topology": "Nabla" is one of the parameters included in the modification command 20 of "move."

[0054] "target":"magnet" specifies the name of the part to be corrected, and in this example, indicates that the name of the part to be corrected is "magnet" (permanent magnet). "target":"magnet" is one of the parameters included in the correction command 20 of "move."

[0055] "Part":"front" specifies the part location to be corrected, and in this example, indicates that the part location to be corrected is the "front" (front side of the permanent magnet). "Part":"front" is one of the parameters included in the "move" correction command 20.

[0056] "Direction": 0 specifies the direction of correction (direction of movement), and in this case, the direction of correction indicates a predetermined direction specified by the numerical value 0. "Direction": 0 is one of the parameters included in the "move" correction command 20.

[0057] "Amount": 2.0 is a specification of the amount of movement (mm), and in this case indicates that the amount of movement is 2.0 mm. "Amount": 2.0 is one of the parameters included in the "move" modification command 20.

[0058] In the shape modification process in response to the modification command 20 in FIG. 4B, the front vertex of the permanent magnet in the rotor topology of "Nabla" is moved 2.0 mm in the direction of 0.

[0059] Returning to FIG. 4A, the "move_single_point" command moves a specific single node in a specified direction and distance. The "rotate" command rotates the entire graph or a specific component by a specified angle around a specified center point. The "designate" command sets a specified dimension size to a specific value and is implemented in combination with the "move" command. The "enlarge" command enlarges or reduces a specified component according to a specified ratio. The center of enlargement or reduction is usually the center point of the component. The "approximate" command approximates a specified component (e.g., a permanent magnet) to a rectangle.

[0060] Returning to FIG. 3, in post-processing S33, the graph information corrected in step S32 is decoded, restored to CAD information, and output as output data 40 (output shape).

[0061] In the shape correction process S32, there is a possibility that the shape may be destroyed by the correction based on the correction command 20. Therefore, the shape correction system 300 executes an error determination (step S34). In the error determination of step S34, two determination processes are executed to determine whether the shape after correction is valid.

[0062] The first determination process (first determination process) is a process for verifying whether each node of the graph is within the valid area of ​​the design. Specifically, the first determination process verifies that the distance from the origin of each node is within the range of the inner and outer diameters of the rotor 12, and that the angle of each node is within the angle of the axis of symmetry. If these cannot be confirmed, an error (shape failure) is determined.

[0063] Another judgment process (second judgment process) verifies whether different components (particularly permanent magnets and flux barriers) in the rotor 12 overlap each other. Specifically, the second judgment process converts the electromagnetic steel sheet portion and other areas into 0 / 1 using image processing technology, performs contour detection, and determines whether the number of detected contours matches the number specified for each topology. If the components overlap each other, it is judged to be an error (a collapse of the shape).

[0064] The design system 100 according to the embodiment is configured as an interactive design system that allows a shape modification system 300, which requires input of modification commands 20 in a predetermined format, to be used in an interactive manner including a user's linguistic instruction 10. For interaction with the user, the design system according to the embodiment includes a command generation system 200. The command generation system 200 generates modification commands 20 for shape modification processing in the shape modification system 300 based on the user's instruction 10. The modification commands 20 are generated, for example, by artificial intelligence (AI). The AI ​​has learned multiple modification commands 20 pre-defined for shape modification in the shape modification system 300, and can select a modification command 20 from the multiple modification commands 20 that corresponds to the linguistic instruction 10 and determine parameters included in the modification command 20 according to the linguistic instruction 10. By learning the multiple modification commands 20, the AI ​​can learn the format and meaning of each of the multiple modification commands 20. As described above, each of the multiple modification commands 20 has a format consisting of a command name (type of modification operation) and one or more predetermined parameters corresponding to the command name. The artificial intelligence can learn the format of each modified command 20 and the meaning of the command name and parameters contained in that format by, for example, providing the artificial intelligence with prompts that explain the format and meaning of the modified command 20.

[0065] The artificial intelligence that generates the correction command 20 may be provided with the linguistic instruction 10 as well as the target data 30 before the shape is corrected. Given the target data 30, the artificial intelligence can understand the shape of the given target data 30 and generate the correction command 20 for appropriately correcting the shape in accordance with the linguistic instruction 10. To facilitate the AI's understanding of the shape of the target data 30 (e.g., a motor shape), the target data 30 may include not only shape data of the target but also explanatory text (description of each part of the target in natural language) indicating each part of the target in the shape data. For example, if the target is a motor, the target data 30 may include explanatory data indicating in language (terms such as stator 11 (electromagnetic steel sheet), rotor 12 (electromagnetic steel sheet), permanent magnet 15, slot 16, and flux barrier (the white part adjacent to the permanent magnet 15 in FIG. 2A ) whether each part of the motor shape is the stator 11 (electromagnetic steel sheet), rotor 12 (electromagnetic steel sheet), permanent magnet 15, slot 16, or flux barrier (the white part adjacent to the permanent magnet 15 in FIG. 2A ) as shown in FIG. 2A ). The AI ​​can understand or grasp each part of the object from this explanation data. Furthermore, the AI ​​that understands or grasps each part of the object can linguistically identify the part that needs to be corrected (for example, the name of the part to be corrected or the part that needs to be corrected in the part to be corrected) based on the linguistic instruction 10 to correct the specific part of the object.

[0066] As described above, the correction command 20 may include parameters for correcting the shape of the target data 30 in accordance with the linguistic instruction 10. The artificial intelligence may determine the parameters to be included in the correction command 20 based on the linguistic instruction 10. The artificial intelligence may determine the parameters based on the linguistic instruction 10 and the target data 30 before correction. The artificial intelligence, which understands or grasps each part of the target based on the target data 30, can identify the name of the part to be corrected or the part of the part to be corrected as a parameter based on the linguistic instruction 10 to correct a specific part of the target. For example, if the linguistic instruction 10 is a correction instruction to move the front position of a permanent magnet, the artificial intelligence can understand the instruction and determine the name of the correction command for the movement ("move") and parameters specifying the target of the movement ("target":"magnet" and "part":"front"). Furthermore, if the linguistic instruction 10 includes an instruction that the movement amount is 2 mm, the artificial intelligence can determine a parameter indicating the movement amount ("amount":2.0). The artificial intelligence also generates other parameters for generating the corrective command 20 based on the linguistic instruction 10 .

[0067] The generation of the modified commands 20 by the command generation system 200 is, for example, performed by a large-scale language model 400 (LLM). The LLM 400 is, for example, GPT-4 from OpenAI. The LLM is an example of generative artificial intelligence (generative AI) that generates language data. The LLM is, for example, configured with a transformer. As shown in FIGS. 5 and 6 , in the embodiment, for example, three independent LLMs are set in the design system 100: a processing judgment LLM (first LLM) 410, a JSON creation LLM (second LLM) 420, and a dialogue LLM (third LLM) 430.

[0068] The processing decision LLM 410 shown in FIG. 5A analyzes a user's input (instruction 10) and determines whether a corresponding command exists. When the processing decision LLM 410 receives a user's linguistic instruction 10 (text instruction), it understands the user's intent and selects and outputs a command name corresponding to the user's instruction 10 from the six commands shown in FIG. 4A. In this way, the processing decision LLM 410 determines the command name (type of correction action) in the correction command based on the instruction 10. If a corresponding command does not exist, it outputs an error. When the user's instruction 10 (prompt) is a prompt unrelated to shape correction, such as "hello," the processing decision LLM 410 enables early error handling before the JSON creation LLM 420 generates a correction command 20. In this way, the processing decision LLM 410 determines whether processing by the next JSON creation LLM 420 is possible.

[0069] The JSON creation LLM 420 shown in FIG. 5B generates a specific JSON-formatted modification command 20 based on user input (instruction 10). The JSON creation LLM 420 creates a detailed JSON-formatted modification command 20 (e.g., the modification command shown in FIG. 4B) from the user's instruction 10 in accordance with the command name (e.g., "move") identified by the processing determination LLM 410. That is, the JSON creation LLM 420 determines parameters to be included in the modification command 20 based on the command name (e.g., "move") and completes the modification command 20. Note that the JSON creation LLM 420 may obtain the command name (e.g., "move") corresponding to instruction 10 from the processing determination LLM 410 as shown in FIG. 5B, or the JSON creation LLM 420 itself may determine the command name based on instruction 10. To determine appropriate parameters, the JSON creation LLM 420 may be provided with the target data 30 before modification. Given the target data 30, the JSON creation LLM 420 can grasp the shape of the provided target data 30 and determine parameters for appropriately modifying the shape in accordance with the instruction 10. The JSON creation LLM 420 also has the function of determining whether the instruction is unclear when determining parameters, for example, when a user instruction 10 to move a magnet does not specify which magnet is being moved. If the JSON creation LLM 420 determines that the instruction 10 does not contain all the information necessary to determine the parameters and that there is missing information, it also has the function of connecting to a process (interactive LLM 430) that prompts the user to confirm the missing information.

[0070] 5C, the dialogue LLM 430 interacts with the user. The dialogue LLM 430 generates a response to an input (instruction 10) from the user, taking into account the outputs of the other two LLMs 410 and 420, thereby realizing natural communication.

[0071] To set up three LLMs 410, 420, and 430 in the LLM 400, role definition statements 510, 520, and 530 for each LLM 410, 420, and 430 are used. Each definition statement 510, 520, and 530 is input as a prompt to the LLM 400 to set up each LLM 410, 420, and 430. By being assigned a role, the LLM 400 can produce output according to that role.

[0072] The definition statement 510 of the role of the processing determination LLM 410 may include a statement defining the role of the LLM 410, such as, for example, "You accept a user instruction to modify the motor's shape, select a command name corresponding to the user instruction from a command list (e.g., the list of modification commands shown in FIG. 4A ), and generate an output." The definition statement 510 may include, for each modification command 20, a statement explaining the output format of the LLM 410, an example of the user instruction 10, and an example of the output to be generated. The definition statement 510 may include a statement explaining the meaning or role of each modification command 20. When the LLM 410 is given a prompt including this definition statement 510, it learns multiple modification commands (e.g., the multiple modification commands shown in FIG. 4A ) pre-defined for shape modification in the shape modification system 300 and operates as a "processing determination LLM 410." When a shape modification instruction 10 is given, the processing determination LLM 410 can select and output the command name of a modification command for shape modification corresponding to the instruction from among the multiple modification command names.

[0073] The role definition statement 520 of the JSON creation LLM 420 may include a statement defining the role of the LLM 420, such as, "You accept user instructions for modifying the motor geometry and generate JSON-formatted output in response to the user instructions." The definition statement 520 may also include a statement instructing how to handle missing information (what to do if necessary information is not available). By providing this statement to the LLM 420, if there is missing information in the user instructions 10, the missing information is indicated in the output generated by the LLM 420.

[0074] The definition statement 520 may include a sentence explaining the command name and parameter meaning of each modification command 20. The definition statement 520 may also include a user instruction 10 and an example of a modification command 20 in JSON format to be generated. This example allows the LLM 420 to understand the context and usage scenario of the command and learn the correct format (e.g., key and value pairs, data type, hierarchical structure, etc.). This is expected to improve the LLM 420's ability to generate accurate output. When the LLM 410 is given a prompt including this definition statement 510, it will have learned multiple modification commands (e.g., the multiple modification commands shown in FIG. 4A) pre-defined for modifying shapes in the shape modification system 300, as well as the format of each modification command 20 (see FIG. 4B), and will operate as a "JSON creation LLM 420." When a shape modification instruction 10 is given, the JSON creation LLM 420 can determine one or more parameters for a modification command for the shape modification in accordance with the instruction and output a completed modification command 20.

[0075] Furthermore, in order to allow the LLMs 410 and 420 to understand the functions of the shape modification system 300, a detailed explanation 550 of the functions of the shape modification system 300 is input to the LLMs 410 and 420 as a prompt. Furthermore, a vocabulary list of technical terms in the motor field is input to the LLMs 410 and 420. The vocabulary list reduces ambiguity in meaning and clarifies different technical terms that have the same concept. This increases flexibility in response to various user inputs and is expected to reduce the probability that the LLMs will output incorrect responses.

[0076] The role definition statement 530 of the interactive LLM 430 can include a statement that generates a response to the user based on the output of the other LLMs 410 and 420, and if there is an output from the LLM 420 indicating missing information, prompts the user to input the missing information.

[0077] 7 shows an example of a procedure for shape modification by the design system 100. When a user inputs a text instruction 10 to the design system 100, the instruction 10 is evaluated by the processing determination LLM 410 (step S71). In step S71, the processing determination LLM 410 analyzes the instruction 10 and determines whether there is a modification command (command name) corresponding to the instruction 10. More specifically, the processing determination LLM 410 determines whether there is a modification command (command name) corresponding to the instruction 10 among the multiple modification commands it has learned (e.g., the multiple modification commands shown in FIG. 4A). If there is a modification command corresponding to the instruction 10, it is possible to process the instruction ("Yes" in step S71), and the process proceeds to step S72.

[0078] If there is no correction command corresponding to instruction 10, processing is not possible ("No" in step S71), and the process proceeds to step S77. In step S77, the interactive LLM 430 creates a response prompting the user to re-input instruction 10. The response is displayed to the user (step S78).

[0079] In step S72, the JSON creation LLM 420 determines parameters for the modification command from the user's text instruction 10 and creates a modification command 20 (modification instruction JSON). In step S73, the JSON creation LLM 420 determines whether the information in the user's instruction 10 is complete (whether all the necessary information is included). If the information is not complete ("No" in step S73), the process proceeds to step S77. In step S77, the interactive LLM 430 creates a response prompting the user to enter the necessary information. The response is displayed to the user (step S78).

[0080] If the necessary information is available ("Yes" in step S73), the process proceeds to step S74.

[0081] In step S74, the shape modification system 300 executes the shape modification process based on the modification command 20. If an error occurs in the shape modification, a response notifying the error is created and displayed (steps S77 and S78).

[0082] When the execution of the shape correction process is completed normally, the interactive LLM 430 creates a response to the user (step S75) based on the instruction 10 and the correction command 20. The response is displayed together with the corrected shape (step S76).

[0083] The design system 100 stores the above-described interaction history between the user and the system 100 in memory (step S79). The user can also provide the system 100 with an instruction 10 to further modify the modified shape. In this case, the system 100 executes the process again from step S71.

[0084] 8 and 9 show examples of user instructions 10 (user input) and the output of each LLM 410, 420, 430. FIG.

[0085] 8 shows "Example 1." In Example 1, the user instruction 10 is "Please set the thickness of the front-side permanent magnet to 6.0 mm." In Example 1, the processing decision LLM (LLM1) 410 outputs {'command': 'designate'}. This output indicates that the command name selected by the processing decision LLM 410 based on the user instruction is "designate."

[0086] In Example 1, the JSON creation LLM 420 determines multiple parameters for the command name: "designate" based on instruction 10, completes the modified command 20 shown in Figure 8, and outputs it. The dialogue LLM 430 also generates and outputs the reply sentence shown in Figure 8 based on instruction 10 and the modified command 20.

[0087] FIG. 9A shows "Example 2." Example 2 illustrates a case in which an error occurs due to shape modification. In Example 2, the user instruction 10 is "Please increase the thickness of the permanent magnet by 1.2 times." In Example 2, the processing decision LLM 410 outputs the command name "enlarge," and the JSON creation LLM 420 outputs the modification command 20, but an error occurs as a result of the shape modification process. In this case, the shape modification system 300 outputs "point idx 1 is out of radial boundary," indicating that the vertex has moved outside the design domain as a result of the shape modification.

[0088] 9B shows "Example 3." In Example 3, the user instruction 10 is "Hello," which is a user input unrelated to shape modification. In this case, the processing determination LLM 410 determines that processing is impossible and creates and displays a response indicating that processing is impossible or a response prompting re-entry of the instruction (steps S71, S77, and S78 in FIG. 7). If the processing determination LLM 410 determines that processing is unnecessary, the JSON creation LLM 420 does not perform processing, thereby avoiding unnecessary processing.

[0089] 10 shows an example of a dialogue between the design system 100 and a user. In the example of the dialogue in Fig. 10, the user provides the system 100 with target data 30 to be modified (step S101) and also provides instructions 10 to the system 100 (step S102). The system 100 executes a shape modification process to modify the shape of the target data 30 based on the instructions 10. The interactive LLM 430 of the system 100 generates and outputs a response to the user (step S103) and also outputs the modified shape (step S104).

[0090] The user issues an instruction 10 to the system 100 to further modify the modified shape (step S105). The system 100 then executes a shape modification process to further modify the modified shape based on the instruction 10. The interactive LLM 430 of the system 100 generates and outputs a response to the user (step S106) and also outputs the further modified shape (step S107).

[0091] Fig. 11 shows a modified example of the design system 100. The design system 100 according to the modified example is configured so that the design system 100 can autonomously execute shape modifications that are repeated multiple times through dialogue with the user as shown in Fig. 10 without the intervention of a human user. This allows the design to be performed autonomously without human intervention.

[0092] The design system 100 shown in Fig. 11 includes an instruction generation system 600 that plays the role of a human user. The instruction generation system 600 is configured using artificial intelligence.

[0093] The instruction generation system 600 is, for example, configured with a large-scale language model 800 (LLM). The LLM 800 is, for example, GPT-4 from OpenAI. The LLM 800 generates linguistic instructions 10 (natural language instructions) for shape modification. The LLM 80, for example, acquires shape-modified output data 40 (which may include explanatory text) from the shape modification system 300, and can generate linguistic instructions 10 for further shape modification using the output data 40 as target data. The linguistic instructions 10 generated by the instruction generation system 600 as a user are provided to the command generation system 200, which generates modification commands 20 based on the linguistic instructions 10. The shape modification system 300 can then re-modify the shape based on the modification commands 20.

[0094] A human user may initially provide linguistic instructions 10A (initial instructions) to the instruction generation system 600. In each of the multiple repeated generation of instructions 10, the instruction generation system 600 may generate new linguistic instructions 10 for appropriately modifying the output data 40 based on the linguistic instructions 10A (initial instructions).

[0095] The instruction generation system 600 may be provided with the corrective command 20 generated by the command generation system 200. In this case, the instruction generation system 600 can generate a new linguistic instruction 10 based on the corrective command 20 up to that point.

[0096] The present invention is not limited to the above-described embodiment, and various modifications are possible.

[0097] 10: Language instruction 10A: Language instruction 11: Stator 12: Rotor 15: Permanent magnet 16: Slot 20: Correction command 30: Target data 40: Output data 100: Design system 200: Command generation system 300: Shape correction system 400: Large-scale language model 410: Processing judgment LLM 420: JSON creation LLM 430: Interactive LLM 510: Definition statement 520: Definition statement 530: Definition statement 550: Detailed explanation statement 600: Instruction generation system 800: Large-scale language model

Claims

1. A design system having a processor that performs operations including: receiving, from a user, linguistic instructions regarding shape modification of target data indicating the shape of the target; generating, from the linguistic instructions, a modification command for a shape modification process that modifies the shape of the target data; and performing the shape modification process of the target data based on the modification command.

2. The design system of claim 1, wherein generating the modification command includes inputting the linguistic instruction into an artificial intelligence and obtaining from the artificial intelligence a modification command for modifying the shape in accordance with the linguistic instruction; and wherein the shape modification process includes modifying the shape of the target data based on the modification command by a shape modification system to which the modification command obtained from the artificial intelligence has been input; and wherein the artificial intelligence has learned a plurality of modification commands that have been preset in the shape modification system for modifying the shape, and selects and outputs a modification command for modifying the shape in accordance with the linguistic instruction from the plurality of modification commands that it has learned.

3. The design system described in claim 2, wherein the artificial intelligence selects a modification command from the plurality of modification commands it has learned to shape the target data before modification in accordance with the linguistic instruction, and determines and outputs parameters for the selected modification command based on the linguistic instruction.

4. The design system according to claim 3, wherein the operation further includes providing the target data before modification to the artificial intelligence, and the artificial intelligence determines the parameters based also on the target data before modification.

5. The design system according to claim 1, wherein the object is a motor, and the modification command is a command for modifying the shape of the object data indicating the shape of the motor.

6. The design system of claim 2, wherein the modification command is a command that conforms to an input format for the shape modification process, the shape modification system is configured to execute the shape modification process by inputting the modification command that conforms to the input format, and the artificial intelligence has learned the input format of the modification command and generates a modification command for shape modification in response to the linguistic instruction in accordance with the input format.

7. The design system of claim 1, wherein the operation further includes: determining whether there is a correction command corresponding to the language instruction; and if it is determined that there is no correction command corresponding to the language instruction, prompting the user to re-input the language instruction; and generating the correction command is executed if it is determined that there is a correction command corresponding to the language instruction.

8. The design system of claim 7, wherein determining whether there is a modification command corresponding to the linguistic instruction includes determining, by an artificial intelligence that has learned a plurality of modification commands for modifying a shape, whether a modification command corresponding to the linguistic instruction is among the plurality of modification commands learned by the artificial intelligence.

9. The design system of claim 1, wherein the operation further includes: determining whether the information necessary for generating the modification command is available in the language instructions; and if it is determined that the information is not available, prompting the user to input the information; and executing the shape modification process if it is determined that the information is available.

10. The design system of claim 9, wherein determining whether the information necessary for generating the modification command is available in the language instructions includes determining whether the information necessary for generating the modification command according to the format is available in the language instructions using artificial intelligence that has learned the format of modification commands for modifying shapes.

11. The design system of claim 10, wherein the modification command according to the format includes parameters for modifying a shape, and determining whether the information necessary for generating the modification command according to the format is available in the language instructions includes determining whether the information necessary for determining the parameters is available in the language instructions.

12. The design system of claim 1, wherein the operation further comprises: determining, by an artificial intelligence that has learned a plurality of modification commands for modifying a shape, whether a modification command corresponding to the linguistic instruction is included in the plurality of modification commands learned by the artificial intelligence; and, if it is determined that a modification command corresponding to the linguistic instruction is not included in the plurality of modification commands learned by the artificial intelligence, prompting the user to re-input the linguistic instruction; generating the modification command is performed if it is determined that a modification command corresponding to the linguistic instruction is included in the plurality of modification commands learned by the artificial intelligence; and generating the modification command includes, by an artificial intelligence that has learned a format of modification commands for modifying a shape, generating a modification command corresponding to the linguistic instruction in accordance with the format; the artificial intelligence that has learned the format of modification commands for modifying a shape determines whether information necessary for generating the modification command in accordance with the format is included in the linguistic instruction; and, if it is determined that the information is not included, prompting the user to input the information.

13. The design system of claim 12, wherein the modification command according to the format includes parameters for modifying a shape, and determining whether the information required to generate the modification command according to the format is available in the language instructions includes determining whether the information required to determine the parameters is available in the language instructions.

14. The design system of claim 1, wherein generating the modification commands from the linguistic instructions is performed by a large-scale language model.

15. A command generation system having a processor that performs operations including: receiving, from a user, linguistic instructions regarding shape modification of object data indicating the shape of the object; and generating, from the linguistic instructions, a modification command for a shape modification process that modifies the shape of the object data.

16. A computer-implemented method executed by a computer, comprising: receiving, from a user, linguistic instructions regarding shape modification of target data indicating a shape of the target; generating, from the linguistic instructions, a modification command for a shape modification process that modifies the shape of the target data; and executing the shape modification process of the target data based on the modification command.

17. A computer program that causes a computer to perform operations including: receiving from a user linguistic instructions regarding shape modification of target data indicating the shape of the target; generating, from said linguistic instructions, a modification command for shape modification processing that modifies the shape of said target data; and executing said shape modification processing of said target data based on said modification command.

Images