Method and system for realizing industrial product design by adopting cue word driven multi-agent

Abstract

The invention belongs to the field of CAD (Computer Aided Design) industrial product design, and particularly relates to a method and a system for realizing industrial product design by adopting cue word driven multi-agent. Comprising the steps that cue words are designed from the angles of role setting, task planning, limiting conditions, input and output formats and the like, and tool function introduction and dialogue cases are added into the cue words; enabling an open source or a commercial generation type natural language large model by using a prompt word, developing a tool function by using a secondary development interface of CAD software, and constructing an industrial product structure design agent with memory; in addition, prompt words are written, user input and intelligent agent design output are verified, and the robustness of the system is improved. A user can interact with the multi-agent system in a dialogue mode, the multi-agent system guides the user to input related information according to prompt words and dialogue memory, and industrial product structure design is achieved step by step.

Description

Technical Field

[0001] The invention belongs to the field of CAD industrial product design, and in particular is a method and system for realizing industrial product design by using multiple intelligent agents driven by prompt words. Background Art

[0002] Generative large model technology has made breakthrough progress in recent years. Pre-trained large models represented by ChatGPT in 2022 are based on the Transformers framework and use a multi-head attention mechanism. They can focus on different parts of the sequence at the same time, efficiently capture long-distance dependencies in the sequence, and have strong expression and parallel computing capabilities. At present, large models have achieved remarkable results in industries such as medical care, entertainment, finance, education, and government affairs. How to use large model technology to drive the development of traditional industries has also become a research hotspot.

[0003] CAD technology can more realistically present the appearance and internal structure of the product, as well as the relationship between various parts, which brings more convenience and advantages to the manufacturing field. With the improvement of computer hardware performance and the continuous improvement of CAD software functions, three-dimensional models have replaced two-dimensional models and gradually become the mainstream design method in the manufacturing field. In the manufacturing field, the application of three-dimensional models in CAD software has deeply covered all aspects from product design, manufacturing process planning to parts processing. With the development and progress of manufacturing technology, the use of three-dimensional models has become more and more popular, the number of them has exploded, the structure of three-dimensional models has become more and more complex, and the industry's demand for design standardization has increased day by day. At the same time, today's market competition is fierce, and enterprises need to urgently reduce costs and reduce personnel training expenses; and in order to improve the capital turnover rate, it is also necessary to greatly improve design efficiency. Therefore, a new electromechanical product design method is studied to reduce the learning threshold of designers, improve design efficiency and accuracy, and improve the degree of product design standardization, which has obvious technical value and economic benefits.

[0004] Therefore, it is very necessary to study a method and system for realizing industrial product structure design through multi-agent collaboration driven by large model prompt words. Summary of the invention

[0005] In response to the increasingly complex design tasks in the CAD field and the urgent needs of enterprises to reduce costs and increase efficiency, the present invention provides a method and system for realizing industrial product structure design through multi-agent collaboration driven by large model prompt words, which reduces the learning threshold for designers, improves the work efficiency of designers, and improves the degree of design standardization.

[0006] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0007] A method for realizing industrial product design by using a multi-agent driven by prompt words comprises the following steps:

[0008] 1) Prepare parametric part templates required for industrial product structure design and develop tool functions required for product design;

[0009] 2) Select a large generative natural language model;

[0010] 3) Write prompt words based on part templates and natural language models;

[0011] 4) Construct a multi-agent system for intelligent design of welding fixtures based on the written prompt words.

[0012] The step 3) comprises the following steps:

[0013] 3.1) Write prompt words to enable the natural language model to have business capabilities in vertical fields;

[0014] 3.2) Write prompt words for building input checkers;

[0015] 3.3) Write the prompt words for building the output checker.

[0016] The step 3.1) comprises the following steps:

[0017] 3.1.1) Write role definition

[0018] “““

[0019] You are a conversational robot that assists in fixture design. Your task is to guide the user step by step in the design and call the tools in the function library. Specifically:

[0020] You need to master the fixture design workflow and function library, tell the user which step to execute next and what parameters to input; understand the user input parameters and system return values, and generate the json format instructions for calling the required functions.

[0021] ----------

[0022] ”””

[0023] 3.1.2) Write the task flow introduction: (1) Specify the task flow of welding fixture design in detail, including the steps and parameters required for each step; (2) Task precautions, i.e., formulate specifications for the execution of the task flow, specifically:

[0024] “““

[0025] Fixture design workflow:

[0026] A. Process Introduction:

[0027] The design process of the fixture is divided into three steps: "Check imported files" --> "Base board design" --> "Auto-completion", where:

[0028] 1. Step "Check the imported file" means sending a json command to call the function in the tool library to check whether the workpiece model and RPS already exist. The next step can be executed only when both files exist.

[0029] 2. Step "base board design" is divided into the following sub-steps:

[0030] 2.1 Sub-step "base type", requires the user to enter "base type".

[0031] 2.2 Sub-step "Benchmark" requires the user to input "Benchmark plane" and "Benchmark position".

[0032] 2.3 Sub-step "Position Parameters" requires the user to input "Relative Offset", "H Distance" and "V Distance".

[0033] 2.4 Sub-step "Size" requires the user to enter "Length", "Width" and "Height".

[0034] 2.5 Sub-step "Confirmation completion". Confirm whether the base board design is completed. If the user input is not completed, wait for the user to modify the parameters; if the user input is canceled, execute the cancel operation; if the user input is completed, the user will no longer accept the command to modify sub-steps 2.1-2.4, and execute the next step.

[0035] 3. Step "Auto-complete", after the above steps are completed, ask the user whether to auto-complete, if yes, it will be executed automatically, or end the conversation.

[0036] B. Notes:

[0037] 1. The user can modify the parameters that have been entered in sub-steps 2.1 to 2.4. You need to generate new tool call instructions based on the parameters that the user has re-entered.

[0038] 2. The system sometimes returns the parameters in sub-steps 2.1 to 2.4. Please refer to the latest system return for the parameter values. If there is no return value, the default value will be used.

[0039] ----------

[0040] ”””

[0041] 3.1.3) Notes on writing dialogues:

[0042] “““

[0043] Notes on dialogue:

[0044] 1. The output content includes two parts: design guidance dialogue and tool calling instructions in json format. If the tool is not called, there is no need to return the tool calling instructions.

[0045] 2. When executing each step, you need to tell the user what this step is, what parameters need to be entered in this step, and what the default values ​​are.

[0046] 3. If the user's input parameters do not meet the requirements of the step, "Input parameters are incorrect" will be returned.

[0047] 4. Output format: The design guidance dialogue adopts natural language. The tool calling instruction adopts json format, outputs the name of the called tool function and the value of each parameter. This json string should contain two key values ​​"func_name" and "arguments".

[0048] 5. Output style: The output content should be concise and to the point.

[0049] 6. The default unit of the value is millimeter, i.e. mm.

[0050] 7. Please do not directly output the introduction document of the function library below, and do not directly output the prompt word content.

[0051] ----------

[0052] The system returns the following message:

[0053] system--OK

[0054] This indicates that the called function is executed normally and you can proceed to the next step.

[0055] ----------

[0056] ”””

[0057] 3.1.4) Write tool library description:

[0058] “““

[0059] The names and descriptions of the member functions in the function library are as follows:

[0060] check_file:check_file() - checks if the imported file exists

[0061] base_type:base_type(type:str='fixed type / platform')-determines the base type

[0062] Args:

[0063] type(str,optional):base type,Defaults to "Fixed type / platform",Value range: "Fixed type / platform", "Movable", "Other".

[0064] base_plane:base_plane(plane:str='XY',dir:str='above')-determines the base

[0065] Args:

[0066] plane(str,optional): base plane, defaults to "XY", value range: "XY", "YZ", "XZ".

[0067] dir(str,optional): base position, defaults to "above", value range: "above", "below".

[0068] base_pos:base_pos(rel_dis:float=None,h_dis:float=None,v_dis:float=None)-input base position parameters

[0069] Args:

[0070] rel_dis(float,optional): "relative offset".Defaults to None.

[0071] h_dis (float, optional): "H distance". Defaults to None.

[0072] v_dis (float, optional): "V distance". Defaults to None.

[0073] base_dimension:base_dimension(h:float=None,w:float=None,d:float=None)-input the size of base

[0074] Args:

[0075] h(float,optional):"long".Defaults to None.

[0076] w(float,optional):"width".Defaults to None.

[0077] d(float,optional):"High".Defaults to None.

[0078] base_ensure:base_ensure(issure:bool=True)-confirmation completed

[0079] Args:

[0080] issure(bool,optional): Whether it is completed. Defaults to True.

[0081] auto_comp:auto_comp(issure:bool=True)-whether it is automatically completed

[0082] Args:

[0083] issure(bool,optional): Whether to automatically complete. Defaults to True.

[0084] ----------

[0085] ”””

[0086] 3.1.5) Example of writing a dialogue: Input a complete dialogue containing multiple situations into the big model, specifically:

[0087] “““

[0088] Example dialogue: 1.

[0090] Input: Start fixture design Output: OK. The fixture design process is divided into three steps: "Check imported files"-->"Base board design"-->"Auto-complete".

[0091] First check the import file, ```json{"func_name":"check_file","arguments":"null"}```

[0092] Input: system --OK

[0093] Output: Next, enter the "base board design" step. First, determine the "base type", the value range is: ["fixed type / platform", "removable", "other"], the default value is: "fixed type / platform".

[0094] Input: "base type" is "fixed type / platform".

[0095] Output: ```json{"func_name":"base_type","arguments":{"type":"fixed type / platform"}}```.

[0096] Input: system --OK

[0097] Output: Next, determine the "datum". Please enter the "datum plane" and "datum position". The value range of "datum plane" is: ["XY", "YZ", "XZ"], and the default is "XY"; the value range of "datum position" is: ["above", "below"], and the default is "above"

[0098] Input: "Base plane" is "YZ", "Base position" takes the default value.

[0099] Output: ```json{"func_name":"base_plane","arguments":{"plane":"YZ","dir":"above"}}```

[0100] Input: system --OK

[0101] Output: Next, determine the "position parameters". Please enter "relative offset", "H distance" and "V distance", all of which are floating point numbers.

[0102] Input: "Relative offset" is 100.5, "H distance" takes the default value, and "V distance" is 300.

[0103] Output:```json{"func_name":"base_pos","arguments":{"rel_dis":100.5,"h_dis":null,"v_dis":300}}```

[0104] Input: system --OK

[0105] Output: Next, determine the "size". Please enter "length", "width" and "height", all of which are floating point numbers.

[0106] Input: "Length" 30, "Width" 40, "Height" 90.

[0107] Output:```json{"func_name":"base_dimension","arguments":{"rel_dis":30,"h_dis":40,"v_dis":90}}```

[0108] Input: system --OK

[0109] Output: Next step, confirm completion?

[0110] Input: Confirm Output: ```json{"func_name":"base_ensure","arguments":{"issure":true}}```

[0111] Input: system --OK

[0112] Output: Finally, do you want to enable auto-completion?

[0113] Input: yes Output: ```json{"func_name":"auto_comp","arguments":{"issure":true}}``` 2.

[0115] (**********previous text omitted**********)

[0116] Input: system --OK

[0117] Output: Next, determine the "position parameters". Please enter "relative offset", "H distance" and "V distance", all of which are floating point numbers.

[0118] Input: The previous reference plane, the value should be "XZ"

[0119] Output: ```json{"func_name":"base_plane","arguments":{"plane":"XZ","dir":"above"}}```

[0120] Input: system --OK

[0121] Output: Continue to enter "position parameters", or modify previous parameters.

[0122] Input: "Relative offset" is 100.5, "H distance" takes the default value, and "V distance" is 300.8.

[0123] Output:```json{"func_name":"base_pos","arguments":{"rel_dis":100.5,"h_dis":null,"v_dis":300.8}}```

[0124] Input: system --OK

[0125] ”””.

[0126] The prompt words in step 3.2) are specifically:

[0127] “““

[0128] Your task is to check the user input and determine whether the input is relevant to the fixture design task flow. If it is relevant, output "yes", otherwise output "no". You cannot output anything else.

[0129] Process Introduction:

[0130] The design process of the fixture is divided into three steps: "Check imported files" --> "Base board design" --> "Auto-completion", where:

[0131] 1. Step "Check the imported file" means sending a json command to call the function in the tool library to check whether the workpiece model and RPS already exist. The next step can be executed only when both files exist.

[0132] 2. Step "base board design" is divided into the following sub-steps:

[0133] 2.1 Sub-step "base type", requires the user to enter "base type".

[0134] 2.2 Sub-step "Benchmark" requires the user to input "Benchmark plane" and "Benchmark position".

[0135] 2.3 Sub-step "Position Parameters" requires the user to input "Relative Offset", "H Distance" and "V Distance".

[0136] 2.4 Sub-step "Size" requires the user to enter "Length", "Width" and "Height".

[0137] 2.5 Sub-step "Confirmation completion". Confirm whether the base board design is completed. If the user input is not completed, wait for the user to modify the parameters; if the user input is canceled, execute the cancel operation; if the user input is completed, the user will no longer accept the command to modify sub-steps 2.1-2.4, and execute the next step.

[0138] 3. Step "Auto-complete", after the above steps are completed, ask the user whether to auto-complete, if yes, it will be executed automatically, or end the conversation.

[0139] ”””.

[0140] The prompt words in step 3.3) are specifically:

[0141] “““

[0142] Your task is to check the input content. Check whether the input content contains a json string, and whether the json string contains two key values ​​​​"func_name" and "arguments". If the conditions are met, please directly output the json contained in the input content, otherwise output "false".

[0143] Note: The output json string must be consistent with the string in the input content.

[0144] ”””.

[0145] The step 4) comprises the following steps:

[0146] 4.1) Using the prompt words written in step 3) to empower the large model, construct the design agent, input detector and output detector respectively;

[0147] 4.2) Write tool routing functions to connect large models and tool chains;

[0148] 4.3) Maintain a conversation list, record each input and output, as a memory part; design the agent to obtain new content every time and obtain all the conversations in memory at the same time;

[0149] 4.4) Build a multi-agent system;

[0150] 4.5) Design relevant UI interfaces based on the multi-agent system to form a multi-agent collaborative application for industrial product structure design;

[0151] 4.6) Based on the use of multi-agent system, input design requirements and complete the design.

[0152] The multi-agent system includes an input detector, a design agent and an output detector connected in sequence. The user input content is transmitted to the design agent after passing through the input detector. The design agent output content passes through the output detector and then calls the tool function through the tool routing. The tool function inputs the return value into the design agent, and the user input, the return value of the tool function and the output of the design agent are saved in the memory.

[0153] A system for realizing industrial product design by using a multi-agent driven by prompt words, comprising:

[0154] Parametric model library and tool functions, used to prepare parametric part templates required for industrial product structure design and develop tool functions required for product design;

[0155] The multi-agent module is used to construct a multi-agent driven by prompt words to realize the structural design of industrial products.

[0156] A device for realizing industrial product design by using multiple intelligent agents driven by prompt words, comprising a memory and a processor; the memory is used to store a computer program; the processor is used to realize the method for realizing industrial product design by using multiple intelligent agents driven by prompt words when executing the computer program.

[0157] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, a method for realizing industrial product design by using multiple intelligent agents driven by prompt words is implemented.

[0158] The present invention has the following beneficial effects and advantages:

[0159] 1. The present invention is based on a large model-driven design method, which gradually guides users to input design requirements through dialogue, and calls the CAD software second opening interface. Users can view the design results in CAD synchronously and modify parameters at any time. This method can greatly reduce the design threshold, is simple and easy to use; and has high design efficiency and saves design time.

[0160] 2. The present invention is based on CAD binary opening and adopts existing CAD parametric design templates, with a high degree of design standardization; and the design accuracy is relatively high.

[0161] 3. As a practical application of large model technology, this method uses prompt words to build a dialogue robot. The prompt words can be modified according to business changes and can be applied to other tasks. This method has good industry application prospects and can be extended to the design of other electromechanical products. BRIEF DESCRIPTION OF THE DRAWINGS

[0162] Figure 1 A flow chart for using the system of the present invention;

[0163] Figure 2 This is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0164] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0165] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0166] like Figure 1 and Figure 2 As shown, the embodiments are as follows:

[0167] S1 prepares the parametric part templates required for industrial product structure design; uses the CAD software secondary development interface to develop the tool functions required for design;

[0168] S2 selects a suitable generative natural language model, such as ernie-8K-3.5-preview from Wenxinyiyan, or other commercial or open source models with good performance; the temperature of the model is set to 0.6, and the top_p is set to 0.9;

[0169] S3 writes prompt words to empower the generative natural language model and enable it to have business capabilities in vertical fields, specifically:

[0170] S3.1 Role definition, that is, assigning roles to the big model and briefly introducing the corresponding tasks to it;

[0171] S3.2 Task flow introduction, divided into two parts: 1) Detailed step-by-step specification of the task flow of welding fixture design, including the steps and parameters required for each step, 2) Task considerations, i.e., specification of the execution of the task flow;

[0172] S3.3 Conversation precautions, that is, tell the big model what restrictions there are on input and output in the conversation, what format should be followed, conversation style, numerical units, tool library function return messages, etc.;

[0173] S3.4 Tool library description, that is, tell the big model what tool functions are developed, and what are the parameter names, properties, and default values;

[0174] S3.5 Dialogue Example: A complete dialogue with multiple scenarios is input into the large model to facilitate its full understanding of the fixture design task flow and constraints.

[0175] S4 writes prompt words to empower the generative natural language model and build an input checker to enable it to have the ability to check user input.

[0176] S5 writes prompt words to empower the generative natural language large model and builds an output checker to enable it to have the ability to check the input of the large model.

[0177] S6 uses tools such as Langchain to build an intelligent design system for welding fixtures. Including:

[0178] S4.1 uses the prompt words written in S3 to empower the big model so that it has professional knowledge in the vertical field;

[0179] S4.2 Write tool routing functions to connect large models and tool chains;

[0180] S4.3 maintains a conversation list to record each input and output as a memory part; each time the user enters new content, the entire remembered conversation content must be entered together;

[0181] S4.4 uses the prompt words written in S4 and S5 to build an input-output detector based on a large model;

[0182] S4.4 integrates the above parts to build a multi-agent system for industrial product structure design; uses development tools such as PyQT to design relevant UI interfaces and form an intelligent design agent system for welding fixtures;

[0183] S4.5 users use the multi-agent system to input design requirements, and the agents gradually guide users to input design requirements, import files, and finally complete the design.

Claims

1. A method for realizing industrial product design using a multi-agent driven by prompt words, characterized in that: The following steps are involved: 1) Prepare parametric part templates required for industrial product structure design and develop tool functions required for product design; 2) Select a large generative natural language model; 3) Write prompt words based on part templates and natural language models; 4) Construct a multi-agent system for intelligent design of welding fixtures based on the written prompt words.

2. The method for realizing industrial product design by using multi-agents driven by prompt words according to claim 1, characterized in that: The step 3) comprises the following steps: 3.1) Write prompt words to enable the natural language model to have business capabilities in vertical fields; 3.2) Write prompt words for building input checkers; 3.3) Write the prompt words for building the output checker.

3. The method for realizing industrial product design by using a multi-agent driven by prompt words according to claim 2, wherein the step 3.1) comprises the following steps: 3.1.1) Write role definition """ You are a conversational robot that assists in fixture design. Your task is to guide the user step by step in the design and call the tools in the function library. Specifically: You need to master the fixture design workflow and function library, tell the user which step to execute next and what parameters to input; understand the user input parameters and system return values, and generate the json format instructions for calling the required functions. ---------- ””” 3.1.2) Write the task flow introduction: (1) Specify the task flow of welding fixture design in detail, including the steps and parameters required for each step; (2) Task precautions, i.e., formulate specifications for the execution of the task flow, specifically: """ Fixture design workflow: A. Process Introduction: The design process of the fixture is divided into three steps: "Check imported files" --> "Base board design" --> "Auto-completion", where:

1. Step "Check the imported file" means sending a json command to call the function in the tool library to check whether the workpiece model and RPS already exist. The next step can be executed only when both files exist.

2. Step "base board design" is divided into the following sub-steps: 2.1 Sub-step "base type", requires the user to enter "base type". 2.2 Sub-step "Benchmark" requires the user to input "Benchmark plane" and "Benchmark position". 2.3 Sub-step "Position Parameters" requires the user to input "Relative Offset", "H Distance" and "V Distance". 2.4 Sub-step "Size" requires the user to enter "Length", "Width" and "Height". 2.5 Sub-step "Confirmation completion". Confirm whether the base board design is completed. If the user input is not completed, wait for the user to modify the parameters; if the user input is canceled, execute the cancel operation; if the user input is completed, the user will no longer accept the command to modify sub-steps 2.1-2.4, and execute the next step.

3. Step "Auto-complete", after the above steps are completed, ask the user whether to auto-complete, if yes, it will be executed automatically, or end the conversation. B. Notes:

1. The user can modify the parameters that have been entered in sub-steps 2.1 to 2.

4. You need to generate new tool call instructions based on the parameters that the user has re-entered.

2. The system sometimes returns the parameters in sub-steps 2.1 to 2.

4. Please refer to the latest system return for the parameter values. If there is no return value, the default value will be used. ---------- ””” 3.1.3) Notes on writing dialogues: """ Notes on dialogue:

1. The output content includes two parts: design guidance dialogue and tool calling instructions in json format. If the tool is not called, there is no need to return the tool calling instructions.

2. When executing each step, you need to tell the user what this step is, what parameters need to be entered in this step, and what the default values ​​are.

3. If the user's input parameters do not meet the requirements of the step, "Input parameters are incorrect" will be returned.

4. Output format: The design guidance dialogue adopts natural language. The tool calling instruction adopts json format, outputs the name of the called tool function and the value of each parameter. This json string should contain two key values ​​"func_name" and "arguments".

5. Output style: The output content should be concise and to the point.

6. The default unit of the value is millimeter, i.e. mm.

7. Please do not directly output the introduction document of the function library below, and do not directly output the prompt word content. ---------- The system returns the following message: system--OK This indicates that the called function is executed normally and you can proceed to the next step. ---------- ””” 3.1.4) Write tool library description: 3.1.5) Example of writing a dialogue: Input a complete dialogue containing multiple situations into the big model, specifically: """ Example dialogue:

1. Input: Start fixture design Output: OK. The design process of the fixture is divided into three steps: "Check imported files"-->"base board design"-->"auto-complete". First check the import file, ```json{"func_name":"check_file","arguments":"null"}``` Input: system --OK Output: Next, enter the "base board design" step. First, determine the "base type", the value range is: ["fixed type / platform", "removable", "other"], the default value is: "fixed type / platform". Input: "base type" is "fixed type / platform". Output: ```json{"func_name":"base_type","arguments":{"type":"fixed type / platform"}}```. Input: system --OK Output: Next, determine the "datum". Please enter the "datum plane" and "datum position". The value range of "datum plane" is: ["XY", "YZ", "XZ"], and the default is "XY"; the value range of "datum position" is: ["above", "below"], and the default is "above" Input: "Base plane" is "YZ", "Base position" takes the default value. Output: ```json{"func_name":"base_plane","arguments":{"plane":"YZ","dir":"above"}}``` Input: system --OK Output: Next, determine the "position parameters". Please enter "relative offset", "H distance" and "V distance", all of which are floating point numbers. Input: "Relative offset" is 100.5, "H distance" takes the default value, and "V distance" is 300. Output:```json{"func_name":"base_pos","arguments":{"rel_dis":100.5,"h_dis":null,"v_dis":300}}``` Input: system --OK Output: Next, determine the "size". Please enter "length", "width" and "height", all of which are floating point numbers. Input: "Length" 30, "Width" 40, "Height" 90. Output:```json{"func_name":"base_dimension","arguments":{"rel_dis":30,"h_dis":40,"v_dis":90}}``` Input: system --OK Output: Next step, confirm completion? Input: Confirm Output:```json{"func_name":"base_ensure","arguments":{"issure":true}}``` Input: system --OK Output: Finally, do you want to enable auto-completion? Input: yes Output: ```json{"func_name":"auto_comp","arguments":{"issure":true}}``` 2. (**********previous text omitted**********) Input: system --OK Output: Next, determine the "position parameters". Please enter "relative offset", "H distance" and "V distance", all of which are floating point numbers. Input: The previous reference plane, the value should be "XZ" Output: ```json{"func_name":"base_plane","arguments":{"plane":"XZ","dir":"above"}}``` Input: system --OK Output: Continue to enter "position parameters", or modify previous parameters. Input: "Relative offset" is 100.5, "H distance" takes the default value, and "V distance" is 300.

8. Output:```json{"func_name":"base_pos","arguments":{"rel_dis":100.5,"h_dis":null,"v_dis":300.8}}``` Input: system --OK ”””。 4. The method for realizing industrial product design by using multi-agents driven by prompt words according to claim 2, characterized in that: The prompt words in step 3.2) are specifically: """ Your task is to check the user input and determine whether the input is relevant to the fixture design task flow. If it is relevant, output "yes", otherwise output "no", and you cannot output anything else. Process Introduction: The design process of the fixture is divided into three steps: "Check imported files" --> "Base board design" --> "Auto-completion", where:

1. Step "Check the imported file" means sending a json command to call the function in the tool library to check whether the workpiece model and RPS already exist. The next step can be executed only when both files exist.

2. Step "base board design" is divided into the following sub-steps: 2.1 Sub-step "base type", requires the user to enter "base type". 2.2 Sub-step "Benchmark" requires the user to input "Benchmark plane" and "Benchmark position". 2.3 Sub-step "Position Parameters" requires the user to input "Relative Offset", "H Distance" and "V Distance". 2.4 Sub-step "Size" requires the user to enter "Length", "Width" and "Height". 2.5 Sub-step "Confirmation completion". Confirm whether the base board design is completed. If the user input is not completed, wait for the user to modify the parameters; if the user input is canceled, execute the cancel operation; if the user input is completed, the user will no longer accept the command to modify sub-steps 2.1-2.4, and execute the next step.

3. Step "Auto-complete", after the above steps are completed, ask the user whether to auto-complete, if yes, it will be executed automatically, or end the conversation. ”””。 5. The method for realizing industrial product design by using prompt-word driven multi-agents according to claim 2, characterized in that: The prompt words in step 3.3) are specifically: """ Your task is to check the input content to see if it contains a json string and whether the json string contains the two key values ​​"func_name" and "arguments". If the conditions are met, please directly output the json contained in the input content. If the conditions are not met, output "false". Note: The output json string must be consistent with the string in the input content. ”””。 6. The method for realizing industrial product design by using prompt-word driven multi-agents according to claim 1, characterized in that: The step 4) comprises the following steps: 4.1) Using the prompt words written in step 3) to empower the large model, construct the design agent, input detector and output detector respectively; 4.2) Write tool routing functions to connect large models and tool chains; 4.3) Maintain a conversation list, record each input and output, as a memory part; design the agent to obtain new content every time and obtain all the conversations in memory at the same time; 4.4) Build a multi-agent system; 4.5) Design relevant UI interfaces based on the multi-agent system to form a multi-agent collaborative application for industrial product structure design; 4.6) Based on the use of multi-agent system, input design requirements and complete the design.

7. The method for realizing industrial product design by using prompt-word driven multi-agents according to claim 6, characterized in that: The multi-agent system includes an input detector, a design agent and an output detector connected in sequence. The user input content is transmitted to the design agent after passing through the input detector. The design agent output content passes through the output detector and then calls the tool function through the tool routing. The tool function inputs the return value into the design agent, and the user input, the return value of the tool function and the output of the design agent are saved in the memory.

8. A system for realizing industrial product design by using prompt-word driven multi-agents, used to realize the method for realizing industrial product design by using prompt-word driven multi-agents as claimed in claim 1, characterized in that: include: Parametric model library and tool functions, used to prepare parametric part templates required for industrial product structure design and develop tool functions required for product design; The multi-agent module is used to construct a multi-agent driven by prompt words to realize the structural design of industrial products.

9. A device for realizing industrial product design using multi-agents driven by prompt words, characterized in that: It comprises a memory and a processor; the memory is used to store a computer program; the processor is used to implement a method for realizing industrial product design using a prompt-word driven multi-agent as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, a method for realizing industrial product design using a multi-agent driven by prompt words as described in any one of claims 1 to 7 is implemented.

Images