Program creation device and computer-readable storage medium

Abstract

A program creation device according to the present disclosure: receives processing content in a multimodal format; analyzes the processing content and a processing condition according to the format of the processing content; creates a prompt including the processing content, the processing condition, and a processing program creation request; transmits the prompt to a generative AI; receives the processing program created by the generative AI; and outputs the processing program created by the generative AI.

Description

Programming device and computer-readable storage medium

[0001] The present disclosure relates to a program creation device and a computer-readable storage medium.

[0002] Conventionally, there exists a technology for generating programming code using artificial intelligence, for example, Patent Document 1.

[0003] Patent No. 7382095

[0004] The machining program for a machine tool is created by creating a three-dimensional model of the workpiece using CAD (Computer Aided Machining). Once the model design is complete, a tool path is generated using CAM (Computer Aided Modeling) and the tool path is converted (post-processed) into a machining program.

[0005] When changing the machining content (machined shape), it is necessary to go through the CAD and CAM processes, which is cumbersome as it requires switching between work on a numerical control device and work on a PC (Personal Computer).

[0006] It is desirable to be able to easily create machining programs.

[0007] A program creation device according to one aspect of the present disclosure includes a reception unit that receives processing content in a multimodal format, a condition analysis unit that analyzes the processing content and processing conditions according to the format of the processing content, a prompt creation unit that creates a prompt including the processing content, the processing conditions, and a request to create a processing program, a transmission unit that transmits the prompt to a generation AI, a reception unit that receives the processing program created by the generation AI, and an output unit that outputs the processing program created by the generation AI.

[0008] FIG. 1 is a block diagram of a program creation device. FIG. 2 is an example of image data showing the details of processing. FIG. 3 is an example of modeling data showing the details of processing. FIG. 4 is a diagram showing an example of a prompt when a user's instruction is in natural language only. FIG. 5 is a diagram showing an example of a prompt when a user's instruction is in natural language and image data. FIG. 6 is a flowchart explaining the operation of a program creation device. FIG. 7 is an example of image data showing marking content. FIG. 8 is an example of modeling data showing marking content. FIG. 9 is a diagram showing an example of a prompt when a user's instruction is in natural language only. FIG. 10 is a diagram showing an example of a prompt when a user's instruction is in natural language and image data. FIG. 11 is an example of a template for accepting changes to a processing program and marking content. FIG. 12 is a diagram showing an example of a changed part of a processing program. FIG. 13 is a diagram showing an example of a prompt. FIG. 14 is a schematic diagram showing changes to a processing program and marking content. FIG. 15 is a hardware configuration diagram of a program creation device.

[0009] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In the following description, components having the same or similar functions will be denoted by the same reference numerals. Duplicate descriptions of those components may be omitted.

[0010] In this application, "based on XX" means "based on at least XX," and includes cases where it is based on other elements in addition to XX. Furthermore, "based on XX" is not limited to cases where XX is used directly, but also includes cases where it is based on XX that has been calculated or processed. "XX" is any element (for example, any information).

[0011] The program creation device 100 according to this embodiment is implemented in, for example, a numerical control device. It can also be implemented on a computer such as a personal computer attached to the numerical control device, a personal computer connected to the numerical control device via a wired or wireless network, a cell computer, a fog computer, or a cloud server.

[0012] The program creation device 100 is connected to an AI server via a network. The AI ​​server may be distributed. The AI ​​server includes a generation AI. When predetermined data is input, the generation AI learns the probability that the response to the predetermined data will be the predetermined data. By learning from a large number of samples, the generation AI comes to output text with a high probability as a response to the predetermined data when the predetermined data is input. The generation AI can be used for purposes such as dialogue, question and answer session, text summarization, text editing, text translation, text conversion, text modification, text optimization, text interpretation / detection, recognition, prediction, judgment, image generation, and comprehensive judgment.

[0013] The generation AI is assumed to have learned data related to the machining program sufficiently to generate the machining program. The data to be learned includes the machining program, manuals, etc.

[0014] 1 is a block diagram of a program creation device 100 according to a first embodiment. The program creation device 100 includes a reception unit 1, a condition analysis unit 2, a prompt creation unit 3, a transmission unit 4, a reception unit 5, an output unit 6, a program update unit 7, and a storage unit 8.

[0015] The reception unit 1 receives a request to create a machining program and machining content. The machining content refers to information that represents the machining shape. The machining content can be specified multimodally. That is, the reception unit 1 receives data in various formats, such as natural language (text), image data, and modeling data. Image data is a bit image that represents an image using pixels. Modeling data is vector data that represents straight lines, curves, etc. by combining points, lines, polygons, etc. Modeling data is created using CAD.

[0016] The condition analysis unit 2 analyzes the current machining conditions and the machining content specified by the user. Machining conditions are necessary for creating a machining program, but are not expressed in a natural language dialogue. Usually, in a dialogue, only the changes are indicated, with the assumptions being shared. There is a possibility that the necessary information will be insufficient to create a machining program from the information obtained in the dialogue. Therefore, the condition analysis unit 2 stores the machining conditions that are expected to be necessary for creating a machining program at the current stage, and adds the machining content specified by the user with a prompt.

[0017] It is assumed that the machining conditions are already set in the numerical control device, machining program, etc. If they are not set, the condition analysis unit 2 may prompt the user to set the machining conditions. The machining conditions are not particularly limited. The machining conditions in this embodiment relate particularly to the machining shape of the workpiece. For example, in the case of drilling, these include the center coordinates and radius of the hole. The machining conditions may also include information about the workpiece. The information about the workpiece includes the shape of the workpiece, the machining surface, and the machining start position.

[0018] The processing content is multimodal. The reception unit 1 receives the processing content in various formats such as natural language, image data, and modeling data.

[0019] The method of condition analysis varies depending on the data format of the processing content. If the processing content is "natural language only," the processing content is text. The reception unit may receive natural language in the form of structured text entered into a template (described later), or may receive text in a style that resembles a conversation with a human (voice spoken by a user) as natural language. In a style that resembles a conversation with a human, a dialogue screen or the like is displayed to receive instructions from the user, such as "Please add a round hole with a radius of XX at the position of XX" or "Please change the corner radius to XX."

[0020] When the user's instructions are "natural language and image data," the processing content is expressed as an image. FIG. 2 is an example of image data indicating the processing content. In this example, a round hole is specified as the processing content. The condition analysis unit 2 may pass the image data directly to the prompt creation unit 3, or may convert information obtainable from the image data into text and pass it to the prompt creation unit 3. By analyzing the image data using technology such as shape recognition, the processing content, such as coordinates and processing shape, can be converted into text. Furthermore, when the image data is a drawing, the processing content, such as dimensions and processing method, is expressed as text, so the text on the drawing may be recognized and the recognized text may be converted into text according to existing drafting rules.

[0021] When the user's instructions are "natural language and modeling data," the processing content is expressed as a combination of points, lines, polygons, etc. Figure 3 shows an example of modeling data showing the processing content. The processing content of the modeling data can also be converted into text using character matching and character recognition, just like image data.

[0022] The prompt generator 3 generates a prompt based on the analysis results of the condition analyzer 2, the processing details entered by the user, and the user's instructions. The contents of the prompt differ depending on the type of data.

[0023] If the user's instructions are "natural language only," a prompt including the current machining conditions and the machining content is created. FIG. 4 is an example of a prompt. The prompt in FIG. 4 includes the current machining conditions and the machining content instructed by the user. The current machining conditions are text such as "Create a round hole," "Depth is XX," "The machining center of the round hole is XX, XX, XX," "The radius is XX," and "The tool is XX." The machining content instructed by the user is text such as "Please add a round hole," "The machining center is XX," and "The radius is XX."

[0024] When the user's instruction is "natural language + image data," the prompt creation unit adds image data to the prompt created when only natural language is used, creating a prompt such as "Please create a machining program that will machine the workpiece so that it has the same shape as the image data," as shown in FIG. 5. Furthermore, machining details determined using image recognition or character recognition may be added to the prompt. For example, "A round hole will be added," "The machining center is XX, and the radius is XX," etc.

[0025] When the user's instruction is "natural language + modeling data," the prompt creation unit 3 adds the modeling data to the prompt created in the case of only natural language, and creates a prompt such as "Please create a machining program that machines the workpiece so that it has the same shape as the modeling data." Furthermore, machining details determined using image recognition or character recognition may be added to the prompt. For example, "A round hole will be added," "The machining center is XX, and the radius is XX," etc.

[0026] The prompt creation unit 3 may add a processing program to the prompt as an example sentence. The example sentence may be a processing program to be changed this time, a processing program created in the past, or the like. A processing program created in the past may be, for example, a processing program with similar processing content. By reading the example sentence, the language model performs in-context learning, which may improve the accuracy of program creation.

[0027] The sending unit 4 sends the created prompt to the AI ​​server. The language model creates a processing program based on the prompt. Note that because the language model is based on statistics, the same processing program is not created each time. Therefore, multiple processing programs may be created and one that operates as expected may be selected.

[0028] The receiving unit 5 receives the machining program created by the language model. The program update unit 7 stores the machining program received by the receiving unit 5 in the memory unit 8. The program update unit 7 determines whether the machining program operates as expected. It also selects a better machining program from among the machining programs that operate as expected. Simulation is an example of a selection method. As a result of the simulation, it is determined whether the machining content is as expected. Simulation may also be used to select a machining program that will speed up the cycle time.

[0029] The operation of the program creation device 100 will be described with reference to the flowchart in Figure 6. The program creation device 100 receives processing content and a processing program creation request (step S1). The processing content is multimodal. The processing content may be in the form of natural language, image data, modeling data, etc.

[0030] The condition analysis unit 2 analyzes the machining content. In analyzing the machining content, first, the current machining conditions are read out (step S2). The machining conditions in this embodiment relate to the machining shape of the workpiece.

[0031] If the processing content is in natural language only (step S3: natural language), the condition analysis unit 2 passes the accepted natural language and the current processing conditions to the prompt creation unit 3. The prompt creation unit 3 creates a prompt including the user's instruction and the current processing conditions (step S4).

[0032] When the processing content is natural language and image data (step S3: image data), the condition analysis unit 2 may determine the processing content using image recognition, character recognition, etc. The condition analysis unit 2 passes the received natural language and image data to the prompt creation unit 3. The prompt creation unit 3 creates a prompt including the user's instruction, the current processing conditions, the image data, the processing content determined from the image data, and the image data (step S5).

[0033] If the processing content is natural language and modeling data (step S3: modeling data), the condition analysis unit 2 may determine the processing content using character matching, character recognition, etc. The condition analysis unit 2 passes the received natural language and modeling data to the prompt creation unit 3. The prompt creation unit 3 creates a prompt including the user's instruction, the current processing conditions, the modeling data, and the processing content determined from the modeling data (step S6).

[0034] The sending unit 4 sends the created prompt to the AI ​​server (step S7). The language model creates a processing program based on the prompt. The receiving unit 5 receives the processing program created by the language model (step S8).

[0035] The program update unit 7 determines whether the created machining program operates as expected (step S9), and stores the created machining program in the storage unit 8 (step S10).

[0036] As described above, the program creation device 100 uses a generation AI to create a machining program for a machine tool. The machining content can be expressed using natural language, image data, modeling data, etc. The program creation device creates a prompt based on the current machining conditions and the natural language, image data, modeling data, etc. that specify the machining content, and causes the language model to create a machining program.

[0037] Users can change the processing content simply by inputting (or speaking) text or specifying image data or modeling data. Because it uses generative AI, processing flexibility is improved compared to creating processing patterns from predetermined characters and motion patterns. Creating a variety of processing programs may also improve cycle time.

[0038] The generation AI may create a processing program from modeling data instead of from text. Modeling data expresses the shape of a character string as a mathematical formula, which may be advantageous for conversion into a processing program.

[0039] The program creation device 100 includes a storage unit 8 that stores previously created machining programs. The program update unit 7 compares previously created machining programs with newly created machining programs and can select a better machining program.

[0040] Additionally, previously created processing programs can be stored and used for learning. By creating a prompt containing the processing program created for processing and loading it into the generation AI, the language model can be trained in-context.

[0041] [Second Embodiment] A program creation device 100 of the second embodiment creates a marking processing program. Marking is a process of inscribing a pattern on the surface of a workpiece using a drill or laser. The reception unit 1 receives a request to create a processing program for executing marking and the marking content. The marking content refers to information that represents the pattern to be marked. The marking content can be specified multimodally. That is, the reception unit 1 receives data in various formats, such as natural language (text), image data, and modeling data. Image data is a bit image that represents an image using pixels. Modeling data is vector data that represents straight lines, curves, etc. by combining points, lines, polygons, etc. The modeling data is created using CAD.

[0042] The condition analysis unit 2 analyzes the current marking conditions and the marking content specified by the user. The marking conditions are conditions that are necessary for creating a machining program but are not expressed in a natural language dialogue. The condition analysis unit 2 stores the marking conditions that are expected to be necessary for creating a machining program at the current stage, and adds the marking content specified by the user using a prompt.

[0043] The marking conditions are assumed to have already been set in the numerical control device, machining program, etc. If they have not been set, the condition analysis unit 2 may prompt the user to set the marking conditions. The marking conditions are not particularly limited. Examples of marking conditions include character size, character spacing, line spacing, font, and character strings to be marked (including special characters). The marking conditions may also include information about the workpiece. The workpiece information includes the shape of the workpiece, the surface to be marked, and the starting position of marking.

[0044] The marking content is multimodal, and the receiving unit 1 receives the marking content in various formats such as natural language, image data, and modeling data.

[0045] The method of condition analysis differs depending on the data format of the marking content. If the marking content is "natural language only," the marking content is text. The reception unit may receive natural language in the form of structured text entered into a template (described below), or may receive text in a style that resembles a conversation with a human (voice spoken by the user) as natural language. In the conversational style, a chat screen or the like is displayed to receive instructions from the user, such as "Please create a processing program that marks today's date as the manufacturing date" or "Please create a processing program that marks the serial number "LLM112."

[0046] When the user's instruction is "natural language and image data," the marking content is expressed as an image. FIG. 7 is an example of image data showing the marking content. In this example, the characters "AB" are specified as the marking content. The condition analysis unit 2 may pass the image data directly to the prompt creation unit. The condition analysis unit 2 may also convert the marking content into text using techniques such as optical character verification (OCV) and optical character recognition (OCR). The condition analysis unit 2 determines the character string to be marked, the font of the character string, etc., and passes the converted marking content to the prompt creation unit 3.

[0047] When the user's instruction is "natural language and modeling data," the marking content is expressed as a combination of points, lines, polygons, etc. Figure 8 shows an example of modeling data showing the marking content. The marking content of the modeling data can also be converted into text using character matching or character recognition, just like image data.

[0048] The prompt generator 3 generates a prompt based on the analysis results of the condition analyzer 2, the markings entered by the user, and the user's instructions. The contents of the prompt differ depending on the type of data.

[0049] If the user's instructions are "natural language only," a prompt is created that includes the current marking conditions and the marking content. FIG. 9 is an example of a prompt. The prompt in FIG. 9 includes the current marking conditions and the marking content specified by the user. The current marking conditions are text such as "Character size XX, character spacing XX, line spacing XX, font XX. The workpiece is XX," and "The equipment to be used is XX." The marking content specified by the user is text such as "Please create a machining program that marks today's date as the manufacturing date," and "Please create a machining program that marks the serial number "LLM112."

[0050] When the user's instruction is "natural language + image data," the prompt creation unit adds image data to the prompt created for the natural language only case, creating a prompt such as "Read the character string in the image data and mark the same character string," as shown in FIG. 10. The content of the marking determined using character matching or character recognition may also be added to the prompt. For example, "Use the font XX" or "The character string is XX."

[0051] When the user's instruction is "natural language + modeling data," the prompt creation unit 3 adds modeling data to the prompt created in the case of natural language only, and creates a prompt such as "Read the character string in the modeling data and mark the same character string." It may also add the content of the marking determined using character matching or character recognition to the prompt. For example, "Use the font XX" or "The character string is XX."

[0052] The prompt creation unit 3 may add example sentences of edit programs to the prompt. The example sentences include edit programs to be changed this time and edit programs created in the past. The edit programs created in the past are, for example, edit programs with similar marking content. The language model performs in-context learning by reading example sentences, which may improve the accuracy of program creation.

[0053] The sending unit 4 sends the created prompt to the AI ​​server. The language model creates a processing program based on the prompt. Note that because the language model is based on statistics, the same processing program is not created each time. Therefore, multiple processing programs may be created and one that operates as expected may be selected.

[0054] The receiving unit 5 receives the machining program created by the language model. The program update unit 7 stores the machining program received by the receiving unit 5 in the memory unit 8. The program update unit 7 determines whether the machining program operates as expected. It also selects a better machining program from among the machining programs that operate as expected. Simulation is one example of a selection method. As a result of the simulation, it is determined whether the marking content is as expected. Simulation may also be used to select a machining program that will speed up the cycle time.

[0055] Marking is a process that uses a drill or laser to create a pattern on the surface of a workpiece. Marking can involve engraving quality control information such as serial numbers and manufacturing dates onto products. Because the content of quality control information changes frequently, creating a processing program using CAD / CAM every time a change occurs is cumbersome. There are also functions that allow you to create marking processing programs without using CAD / CAM. However, with such functions, processing programs are created from predetermined characters and motion patterns, which limits processing flexibility and leaves room for cycle time improvement.

[0056] In the program creation device 100 of the second embodiment, a machining program is created using generation AI, so a marking machining program can be created without using CAD / CAM. Also, because the machining program is not created from a predetermined operation pattern, the machining program becomes more flexible, which may improve the cycle time.

[0057] [Third Embodiment] In the program creation device 100 of the third embodiment, the condition analysis unit 2 identifies the parts of the processing content that need to be changed and creates a processing program for only the identified changed parts. The changed parts of the processing content are mainly patterns such as character strings. The shape of the workpiece and the position of the processing surface often do not need to be changed. Only the parts that need to be changed are selected, while parts that do not need to be changed are left as they are. Note that the program creation device 100 of the third embodiment has substantially the same configuration as the program creation device of the first embodiment, so only the differences will be described. Furthermore, although the third embodiment will be described using marking processing as an example, it can also be applied to other processing.

[0058] The reception unit 1 of the third embodiment uses a template. FIG. 11 is an example of a template. In the template, the machining program to be changed and the change in marking content can be specified. In the template of FIG. 11, the machining program to be changed is "1", and the marking content is specified to be changed from the letter "B" to the letter "C".

[0059] The condition analysis unit 2 acquires the machining program before the change and identifies the changed parts of the machining program. An example of the changed parts is shown in FIG. 12. In the example of FIG. 12, the program "G01 X100 Q100; G01 Y0.1 Q0; G01 X0. Q100;" is identified. The changed parts are the parts including the marking process of the letter "B." Note that the machining path is not independent for each letter, so parts of adjacent letters may also be included in the changed parts as necessary.

[0060] The prompt creation unit 3 creates a prompt based on the change location read by the condition analysis unit 2 and the marking content specified by the user. FIG. 13 is an example of a prompt. The prompt in FIG. 13 defines the natural language "Please update the machining program," a fragment of the machining program to be changed, and the marking content. The marking content includes information about the characters to be changed ("Before change: 'B'," "After change: 'C'") and the marking conditions. The marking conditions include the type of machining "marking processing," the type of machine tool "laser processing machine," etc.

[0061] In the above example, the marking content is received as text, but the marking content may also be received as image data or modeling data. The condition analysis unit 2 can determine the marking content using techniques such as character matching and character recognition. The condition analysis unit 2 identifies the parts of the machining program that need to be changed based on the analysis results.

[0062] In addition, the content of markings received in text format can be reflected in the modeling data. Generative AI has technology for converting text into modeling data (e.g., Text to CAD). Using such technology, user instructions in text format can be reflected in image data or modeling data.

[0063] The receiving unit 5 receives the processing program created by the language model. An example of the processing program is shown in FIG. 14. The changed part is a part of the processing program. When an existing processing program is updated with the processing program created by the language model, the character string to be marked is changed from "AB" to "AC."

[0064] The output unit 6 replaces a part of the existing machining program with the machining program created by the language model, and creates a new machining program.

[0065] The program update unit 7 stores the created machining program in the storage unit 8. The created machining program includes a program fragment created by a language model and a new machining program that replaces part of an existing machining program.

[0066] As described above, the program creation device 100 of the third embodiment identifies the change locations in the marking content and causes the generation AI to create a machining program for only the identified locations. This shortens the length of the generated machining program and improves accuracy. Furthermore, by limiting the change locations, the load on the generation AI is also reduced. By limiting the change locations, the machining program can also be checked quickly.

[0067] The hardware configuration of a program creation device 100 to which the present disclosure is applied will be described below. Fig. 15 is a hardware configuration diagram of the program creation device 100. As shown in Fig. 15, the program creation device 100 includes a CPU 111 that controls the entire program creation device 100, a ROM 112 that records programs and data, and a RAM 113 for temporarily expanding data. The CPU 111 reads out a system program recorded in the ROM 112 via a bus and determines the consistency between a block and a tool.

[0068] The nonvolatile memory 114 is backed up by, for example, a battery (not shown), and the stored state is maintained even when the power to the program creation device 100 is turned off. The nonvolatile memory 114 stores various data such as programs read from the external device 120 via the interfaces 115, 118, and 119 and operation inputs input via the input device 20. The nonvolatile memory 114 may store programs and data for executing the program creation device 100 of this embodiment.

[0069] The interface 115 is an interface for connecting the program creation device 100 to an external device 120 such as an adapter. Programs, various parameters, etc. are loaded from the external device 120. The interface 118 is an interface for connecting the program creation device 100 to a display device 30 such as a liquid crystal display. The display device 30 displays various data loaded into memory, data obtained as a result of executing programs, etc. The interface 119 is an interface for connecting the program creation device 100 to an input device 20 such as a keyboard or pointing device. The input device 20 passes commands, data, etc. based on operations by an operator to the CPU 111 via the interface 119.

[0070] Although the present disclosure has been described in detail, the present disclosure is not limited to the individual embodiments described above. Various additions, substitutions, modifications, partial deletions, etc. are possible in these embodiments without departing from the gist of the present disclosure or the gist of the present disclosure derived from the claims and their equivalents. Furthermore, these embodiments can also be implemented in combination. For example, in the above-described embodiments, the order of each operation and the order of each process are shown as examples and are not limited to these.

[0071] The following are supplementary notes related to embodiments of the present disclosure. (Supplementary Note 1) The program creation device (100) includes a reception unit (1) that receives processing content in a multimodal format, a condition analysis unit (2) that analyzes the processing content and processing conditions according to the format of the processing content, a prompt creation unit (3) that creates a prompt including the processing content, the processing conditions, and a processing program creation request, a transmission unit (4) that transmits the prompt to a generation AI, a reception unit (5) that receives the processing program created by the generation AI, and an output unit (6) that outputs the processing program created by the generation AI. (Supplementary Note 2) The output unit (6) includes a program update unit (7) that stores the processing program created by the generation AI in a memory unit. (Supplementary Note 3) The reception unit (1) receives a request to change the processing content of an existing processing program, and the condition analysis unit (2) identifies the change portion of the processing content in the existing processing program. (Supplementary Note 4) The prompt creation unit (3) creates a prompt including the change portion of the processing program. (Supplementary Note 5) The receiving unit (5) receives a processing program that changes the changes to processing content instructed by the user, and the output unit (6) replaces the changes with the received processing program to create a new processing program. (Supplementary Note 6) The multimodal format is at least one of natural language, image data, and modeling data. (Supplementary Note 7) The condition analysis unit (2) converts the processing content expressed in image data or modeling data into text. (Supplementary Note 8) The prompt creation unit (3) includes previously created processing programs as example sentences in the prompt. (Supplementary Note 9) The computer-readable storage medium (112, 113, 114) stores instructions to cause one or more processors (111) to execute the following processes: accept processing content in a multimodal format, analyze marking content and marking conditions according to the format of the processing content, create a prompt including the processing content, the processing conditions, and a processing program creation request, send the prompt to a generation AI, receive the processing program created by the generation AI, and output the processing program created by the generation AI.

[0072] REFERENCE SIGNS LIST 100 Program creation device 1 Reception unit 2 Condition analysis unit 3 Prompt creation unit 4 Transmission unit 5 Reception unit 6 Output unit 7 Program update unit 8 Storage unit 111 CPU 112 ROM 113 RAM 114 Non-volatile memory

Claims

1. A program creation device comprising: a reception unit that receives processing content in a multimodal format; a condition analysis unit that analyzes the processing content and processing conditions according to the format of the processing content; a prompt creation unit that creates a prompt including the processing content, processing conditions, and a request to create a processing program; a transmission unit that transmits the prompt to a generation AI; a reception unit that receives the processing program created by the generation AI; and an output unit that outputs the processing program created by the generation AI.

2. A program creation device according to claim 1, wherein the output unit includes a program update unit that stores the machining program created by the generation AI in a memory unit.

3. The program creation device according to claim 1, wherein the reception unit receives a request to change the machining content of an existing machining program, and the condition analysis unit identifies the portion of the machining content to be changed in the existing machining program.

4. The program creation device according to claim 3, wherein said prompt creation section creates a prompt including a change in a machining program.

5. A program creation device as described in claim 3, wherein the receiving unit receives a machining program that changes the change portion to the machining content instructed by the user, and the output unit replaces the change portion with the received machining program to create a new machining program.

6. The program creation device according to claim 1, wherein the multimodal format is at least one of natural language, image data, and modeling data.

7. The program creation device according to claim 1, wherein said condition analysis section converts the processing details expressed by image data or modeling data into text.

8. The program creation device according to claim 1, wherein the prompt creation section includes previously created machining programs as example sentences in the prompt.

9. A computer-readable storage medium storing instructions for causing one or more processors to execute the following process: accept processing content in a multimodal format; analyze the processing content and processing conditions according to the format of the processing content; create a prompt including the processing content, processing conditions, and a processing program creation request; send the prompt to a generating AI; receive the processing program created by the generating AI; and output the processing program created by the generating AI.

Images