A human-machine collaborative concept design generation method and system based on FBS theory

Abstract

The present invention discloses a human-machine collaborative conceptual design generation method based on FBS theory, comprising the following steps: obtaining a conceptual design case and extracting keywords based on design requirements and the functions, behaviors and structures in the FBS theory to construct a functional space, a behavioral space and a structural space; embedding the functional space, the behavioral space and the structural space into a large language model framework in sequence based on the logical order of the conceptual design process to construct a conceptual design auxiliary model; inputting the design requirements into the conceptual design auxiliary model to obtain a function-structure mind map. The present invention also provides a human-machine collaborative conceptual design generation system. The method provided by the present invention provides guidance for the designer's thinking process in a structured manner, thereby expanding the design ideas and improving the quality of the final generated design solution.

Description

Technical Field

[0001] The present invention relates to the field of computer-aided design, and in particular to a method and system for generating human-computer collaborative concept design based on FBS theory. Background Art

[0002] Concept design is a key stage in the product development process. In the process of converting design requirements into design goals and solutions, design ideas are generated and transformed into preliminary designs. Although traditional design methods, such as brainstorming and mind mapping, are still effective in some situations, they are very challenging for novice designers, especially when facing more complex problems, because they lack the key design knowledge that experienced designers have. This review of relevant methodological research in the field of conceptual design includes not only the optimization of design processes and design methods, but also the exploration of using computer-aided means to help designers generate design ideas.

[0003] Concept design is a time-consuming, labor-intensive and complex process. A systematic and comprehensive concept design method, supported by big data retrieval, can help designers thoroughly explore the entire space of options in the early stages of product development, and greatly reduce the time and economic cost of trial and error in the later stages of the development process, reducing the possibility of design failure or being surpassed by competing products.

[0004] In order to discover the rules in conceptual design and represent the conceptual design process in a general way, Gero proposed the FBS (Function-Behavior-Structure) model.

[0005] The model decomposes the design task into a hierarchical system consisting of functions, behaviors, and structures, providing a task decomposition method from fuzzy requirements to potential structures.

[0006] Patent document CN114399277A provides a method and system for assisting concept designers in custom product concept design. The system consists of three main modules: a knowledge push module, an information interaction module, and a process design module. The knowledge push module is responsible for providing design resources and retrieving relevant knowledge; the information interaction module provides various personnel with the function of inputting design information and allows them to select the category of design information; the process design module provides a solution creation window, establishes a process solution in the form of a structure diagram, and supports concept designers to edit it. Through the collaborative work of these modules, the invention aims to assist designers in product innovation design in a structured manner. However, since no comparative experiments are carried out in the patent, considering that an overly structured and complex method may have problems in practical application, the operability and effectiveness of the method remain to be proven. In addition, the process design module of the design method combines a large number of tools, such as applying FBS theory to product functional design and working principle scheme design. Although FBS theory has been proven to be an important tool for cross-domain representation and analysis of complex design problems, it relies more on the user's personal knowledge accumulation when obtaining inspiration concepts and exploring potential structures, which makes it impossible to guarantee high-quality design solutions, especially for inexperienced novice designers.

[0007] Patent document CN114357121A proposes a data-driven innovation solution design method and system. The method first determines the innovation concept text and knowledge text database, and identifies the feature clusters in them. Then, the vector of the text is determined based on the feature cluster, and the similarity value between the knowledge text and the innovation concept text is calculated. Finally, the recommended text is selected from the knowledge text database based on the similarity value to determine the innovation solution. This method can recommend knowledge related to the semantics of the innovation concept and assist designers in converting the innovation concept into a design solution. However, the system presents stimuli in the original form of knowledge, and cannot guide users on how to apply inspirational stimuli to design goals. Designers still bear greater mental pressure when generating innovative solutions. Summary of the invention

[0008] The purpose of the present invention is to provide a human-machine collaborative concept design generation method and system based on FBS theory, which can provide guidance for the designer's thinking process in a structured manner, thereby expanding design ideas.

[0009] In order to achieve the first object of the present invention, a human-machine collaborative concept design generation method based on FBS theory is provided, comprising the following steps:

[0010] Obtain conceptual design cases and extract keywords based on design requirements and functions, behaviors, and structures in FBS theory to construct a function space for mapping requirements to functions, a behavior space for mapping functions to behaviors, and a structure space for mapping behaviors to structures;

[0011] Based on the logical order of the conceptual design process, the function space, the behavior space and the structure space are sequentially embedded in the large language model framework to construct a conceptual design auxiliary model, the conceptual design auxiliary model includes a prompt word guidance module, a function solving module, a behavior solving module, a structure solving module and a design scheme summary module, the prompt word guidance module obtains corresponding keywords through a prompt word template, the function solving module generates a corresponding function set according to the keywords of the design requirements, the behavior solving module generates a corresponding behavior set according to the keywords of the functions in the function set, the structure solving module generates a corresponding structure set according to the keywords of the behaviors in the behavior set, the design scheme summary module autonomously derives the behaviors corresponding to each structure according to the obtained structure set by manual means, and compares the consistency with the behavior set output by the behavior solving module, if the derived behavior does not appear in the behavior set, then the structure corresponding to the structure set is eliminated, otherwise the corresponding structure is retained to obtain a structure set with correlation, the functions corresponding to the structure set with correlation are paired to output a visual function-structure mind map;

[0012] The design requirements are input into the conceptual design auxiliary model to obtain the function-structure mind map corresponding to the design requirements, thereby providing designers with diverse, novel and highly usable inspiration stimulation.

[0013] The present invention divides abstract and fuzzy design requirements into three stages for step-by-step reasoning, including converting design requirements into functional space (requirement->function), converting functions into expected behavior space (function->behavior), and generating structures based on expected behaviors (behavior->structure). At the same time, an input template for problems or sub-problems is designed. By calling the interface of the large language model, human-computer collaborative conceptual design based on FBS theory can be realized, helping designers to produce higher-quality conceptual design solutions.

[0014] Specifically, the prompt word guidance module decomposes the design reasoning process into three sub-steps based on the function, behavior and structure in the FBS theory to construct the corresponding prompt word template;

[0015] The prompt word template includes three sub-templates, which are respectively used for solving function space, solving behavior space and solving structure space.

[0016] Specifically, the prompt word template includes a corresponding standard identifier, and the standard identifier includes a function identifier for functional solution, a performance identifier for behavioral solution, and a structure identifier for structural solution.

[0017] Specifically, the expression of the prompt word guidance module is:

[0018] GC=M ptl (PT(x))

[0019] Among them, GC represents the generated text content, x represents the input keyword or phrase, PT represents the prompt template used to package the input content, and M ptl Stands for Large Language Model Framework.

[0020] Specifically, the large language model framework is constructed using a question-answering model and a text generation model.

[0021] Specifically, the hyperparameters of the large language model framework are set as follows:

[0022] The temperature is set to 0.7;

[0023] The maximum length of the output text is set to 256;

[0024] The Top P value is set to 1;

[0025] Frequency is set to 0;

[0026] The presence penalty is set to 0.

[0027] Specifically, the summary module is provided with a screening step in the process of matching functions and structures, and the specific process is as follows:

[0028] Based on the acquired structure set, the behaviors corresponding to each structure are derived manually and compared for consistency with the behavior set output by the behavior solving module;

[0029] If the derived behavior does not appear in the behavior set, the corresponding structure of the structure set is removed, otherwise the corresponding structure is retained to obtain a related structure set.

[0030] Specifically, the function-structure mind map represents the set of implementation structures corresponding to each key function in the design solution in a visual form.

[0031] In order to achieve the second object of the present invention, a human-machine collaborative concept design generation system is provided, which is implemented by the human-machine collaborative concept design generation method based on the FBS theory, and includes a data interaction unit, a data processing unit and a data visualization unit;

[0032] In the data interaction unit, the user can input design requirements, screen and further improve the functions, behaviors and structures generated by the model, and add the improved results to the designed prompt word template;

[0033] The data processing unit, according to the screening results of the data interaction unit, gradually infers the functions, behaviors and structures corresponding to the design requirements, and finally outputs a function-structure set that conforms to the FBS method and can break the designer's mindset in combination with user evaluation and screening;

[0034] The visualization unit generates a corresponding visualization mind map according to the obtained function-structure set.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The human-machine collaborative conceptual design method that combines the large language model with FBS theory helps to make up for individual cognitive limitations by entrusting part of the reasoning task to the large language model with general knowledge, thereby significantly improving the design efficiency and the novelty and diversity of the design schemes.

[0037] At the same time, compared with the traditional generative design method based on the large language model, this method is based on the traditional design method FBS theory for process design, which is more in line with the designer's thinking and cognition. At the same time, designers can filter the generated content at each generation stage, which improves the designer's controllability and helps the large language model output solutions that better meet the designer's needs in subsequent interactions. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart of the method for generating a human-machine collaborative conceptual design provided in this embodiment;

[0039] Figure 2 A diagram of the human-computer interaction collaboration mechanism of the conceptual design auxiliary model provided for this embodiment. DETAILED DESCRIPTION

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0041] like Figure 1 As shown in FIG. 1 , a human-machine collaborative concept design generation method based on FBS theory is provided in this example, which includes the following steps:

[0042] Obtain conceptual design cases and extract keywords based on design requirements and functions, behaviors, and structures in FBS theory to construct a function space for mapping requirements to functions, a behavior space for mapping functions to behaviors, and a structure space for mapping behaviors to structures;

[0043] Based on the logical order of the conceptual design process, the function space, the behavior space and the structure space are sequentially embedded in the large language model framework to construct a conceptual design auxiliary model, the conceptual design auxiliary model includes a prompt word guidance module, a function solving module, a behavior solving module, a structure solving module and a design scheme summary module, the prompt word guidance module obtains corresponding keywords through a prompt word template, the function solving module generates a corresponding function set according to the keywords of the design requirements, the behavior solving module generates a corresponding behavior set according to the keywords of the functions in the function set, the structure solving module generates a corresponding structure set according to the keywords of the behaviors in the behavior set, the design scheme summary module autonomously derives the behaviors corresponding to each structure according to the obtained structure set by manual means, and compares the consistency with the behavior set output by the behavior solving module, if the derived behavior does not appear in the behavior set, then the structure corresponding to the structure set is eliminated, otherwise the corresponding structure is retained to obtain a structure set with correlation, the functions corresponding to the structure set with correlation are paired to output a visual function-structure mind map;

[0044] Input the design requirements into the conceptual design auxiliary model to obtain the function-structure mind map corresponding to the design requirements.

[0045] More specifically, inspired by the FBS theory, the reasoning process is decomposed into three stages, gradually transforming abstract and vague design requirements into design solutions that meet the needs of designers, including transforming design requirements into functional space (requirements->functions), transforming functions into expected behavior space (functions->behaviors), and generating structure space based on expected behaviors (behaviors->structures).

[0046] When determining the interaction mechanism between the large language model and humans, the present invention considers the advantages of both to achieve complementary advantages. On the one hand, the large language model has a wider knowledge space than human individuals and is good at reasoning tasks; on the other hand, the design output depends on human preferences, choices and evaluations.

[0047] like Figure 2As shown, the construction and conversion phases of the FBS ontology are integrated into the large language model reasoning, while the phases from structure to behavior derived from the structure (structure->behavior(derived)), the evaluation phase between expected behavior and derived behavior (behavior(expected)<->behavior(derived)), and the document phase of generating design descriptions based on the structure will be performed by human designers.

[0048] Among them, the behavior (expected) <-> behavior (derived) stage represents the human evaluation of the consistency between behavior (expected) and behavior (derived). Based on the structure generated by the large language model (such as "using a force sensor"), the designer infers the behavior that this structure may exhibit (such as "being able to use a force sensor, that is, behavior (derived)"), and then evaluates the consistency between behavior (derived) and behavior (expected) to help users screen feasible structural design solutions.

[0049] After selecting a structure that can express the function well, the function and the structure that supports this function are combined into a "function-structure" pair. Because when actually conducting inspiration stimulation, it is obvious that there is no need to express the behavioral component as an intermediate medium in the design description. Instead, the function and the corresponding structure reflect the possible design path of the requirement. Therefore, the functional content generated in the "requirement->function" mapping and the corresponding structural content generated in the "behavior->structure" mapping are organized into paired "function-structure" stimuli to help designers better understand the relationship between function and structure in FBS theory.

[0050] In addition, the large language model is trained on massive amounts of text data and has the ability to perform a variety of tasks, such as sentence completion, summaries and keywords, question and answer, and code explanation. Considering that the focus of this application scenario is on the ability of the large language model to make appropriate inferences based on the context, this embodiment specifies question and answer as the inference mode. Based on the designed prompt word template or sub-template, the input is completed, and by calling the interface of the large language model, novel, diverse, and highly usable answers can be generated.

[0051] The existing large language models have shown strong performance in zero-shot learning. Therefore, no further fine-tuning or few-shot learning is performed on the large language models, thus saving the time and labor cost of collecting and building the design dataset.

[0052] In addition, it is a challenging task to directly define a unified large language model prompt word template suitable for FBS. To address this problem, this embodiment decomposes the task prompt words into different sub-prompts. Studies have shown that the method of decomposing prompts in large language models can better handle the problem of large tasks that require multiple predictions for a single sample. However, unlike the prompt word design based on fine-tuning or few-shot learning, the large language model based on zero-shot learning has limited ability to understand the abstract description of the design task and has difficulty analyzing the possible forms of task specifications for ambiguous problems.

[0053] To solve this problem, this embodiment designs an "identifier specification" for each sub-prompt template. An identifier is a keyword or phrase that specifies the information generated by the large language model related to the request target. It can be a noun or a phrase, or a verb that describes the context. The standardized identifier helps the model to reasonably generate answers for each subtask, effectively reducing the difficulty of understanding the prompts for the large language model based on zero-shot learning, making the generated content more reliable.

[0054] The large language model prompt word template proposed in this embodiment is intended to improve the accuracy and controllability of the text generated by the large language model. Existing large language models usually generate answers based on text prompts, and users guide the model to generate the required output by creating appropriate prompt words. However, since prompt words usually require users to edit and input them themselves, if the prompt words are used improperly, it may affect the subsequent links such as encoding, decoding, retrieval and generation by the large language model, resulting in low usability of the generated results.

[0055] In order to enable the large language model to better understand user input and make reasonable inferences, the present invention proposes a design method for a prompt word template. The prompt word template can be regarded as a sentence wrapper used to organize keywords or phrases, and its expression is as follows:

[0056] GC=M ptl (PT(x))

[0057] Among them, GC represents the generated text content, x represents the input keyword or phrase, PT represents the prompt template used to package the input content, and M ptl Represents a large language model used to reason about design examples.

[0058] Its actual visualization is as follows: “What is ( x )”, where [object name]∈{“gear”, “mixer”, “torque”.} represents an optional list of keywords or phrases from which the designer can select an item and fill it into ( x ). The filled-in and organized prompts are then fed into a large language model to generate the corresponding text content.

[0059] In this embodiment, in order to reduce the impact of different hyperparameters and models on different user groups and unify the generation ability of the model, this embodiment adopts the question-answering mode and the text generation model. The hyperparameters set in the experiment are as follows: Temperature is 0.7, the maximum length of the output text is 256, the Top P value is 1, the frequency is 0, and the presence penalty is 0.

[0060] Since the method proposed in this embodiment needs to perform reasoning in the three stages of FBS theory, the experiment designed different prompt word templates for each subtask, as shown in Table 1.

[0061] Table 1

[0062]

[0063] Each prompt word template is constructed as a simple question sentence corresponding to a specific subtask in the FBS process. The canonical identifiers of each template are represented as "function", "expression", and "structure" respectively, so that the model can efficiently capture FBS-related knowledge in the zero-shot learning prompt setting. The placeholders used in the templates, such as "[design object and requirements]", "[function]", and "[behavior]", represent the information that users need to fill in.

[0064] Among them, in the prompt word template of "Function->Behavior", the design concept "behavior" is not used as a standard identifier. This is because the behavior itself is not the implementation carrier of the function, but serves as a logical trigger to trigger the implementation of the function. From the perspective of natural language understanding, if "behavior" is used to let the model infer behavior, the model often assumes that "behavior" is an action that may occur in a general situation. Considering the limited reasoning ability of the model, using the word "expression" makes it easier for the model to understand that the "behavior" ontology is an interactive logic that can provide a certain function. Table 1 shows the prompt word template and identifier specifications.

[0065] In the process of setting the model parameters, in order to reduce the impact of different hyperparameters and models on different user groups and unify the generation capability of the model, this embodiment adopts the question-answering mode and the text generation model. The hyperparameters set in the experiment are as follows: Temperature is 0.7, the maximum length of the output text is 256, the Top P value is 1, the frequency is 0, and the presence penalty is 0.

[0066] In this way, the impact of hyperparameters and model types on different user groups can be reduced, making the generated content more consistent and controllable.

[0067] At the same time, it is worth noting that the theory and method proposed in the present invention are applicable to all large language models of text generation.

[0068] This embodiment also provides a human-machine collaborative concept design generation system, which is implemented by the human-machine collaborative concept design generation method based on FBS theory provided in the above embodiment, including a data interaction unit, a data processing unit and a data visualization unit;

[0069] In the data interaction unit, the user can input design requirements, screen and further improve the functions, behaviors and structures generated by the model, and add the improved results to the designed prompt word template;

[0070] The data processing unit, according to the screening results of the data interaction unit, gradually infers the functions, behaviors and structures corresponding to the design requirements, and finally outputs a function-structure set that conforms to the FBS method and can break the designer's mindset in combination with user evaluation and screening;

[0071] The visualization unit generates a corresponding visualization mind map according to the obtained function-structure set.

[0072] In order to better illustrate the technical effect of the method provided in this embodiment, the experiment recruited 30 designers with similar hand-drawing skills to participate in the design challenge, including 15 novice designers with 3 to 4 years of undergraduate design education background and 15 professional designers with more than 8 years of design experience. Participants need to draw a sketch of a leg exercise rehabilitation device based on the context generated by the large language model.

[0073] The experiment set up three different groups with the goal of evaluating the differences in the effectiveness of large language model generation solutions when using different methods. The experiment was conducted in the same room and at the same time. Each participant was provided with a computer configured and familiarized with how to use their group's method for design generation before the experiment. Each participant first used the large language model to generate a design solution, and then completed design ideas and sketches on the solution. The entire process was limited to 30 minutes, and no other software or search engines were allowed to be used, nor was any content entered beyond the request.

[0074] 1) Control group G1, generates design proposals based on fixed prompt words.

[0075] First, input some “few-shot prompt” examples into the model, such as “Keywords: pet wheelchair, elderly dog; Description: The design of pet wheelchairs solves these drawbacks with airbags…letting them enjoy the rest of their lives.”.

[0076] Then, the keywords of the design requirements are sorted and input into the large language model, such as "Keywords: legs, sports rehabilitation equipment" in this experiment. At this time, the question-answering mode of the large language model imitates the previously learned mode and directly generates the aspect description of the concept design, "The leg sports rehabilitation device is characterized by an adjustable intensity setting, so that the user can customize the massage intensity according to their needs..."

[0077] The designer reads and understands the generated results and screens them. If he is not satisfied with the generated results, he can command the large language model to regenerate them.

[0078] 2) Experimental group G2, input was based on the semi-structured prompt word template of FBS theory.

[0079] First, prepare a “prompt word template”, such as “What functions should a [design object and requirement] have?”, “What kind of design can demonstrate [function]”, “What kind of structural design can achieve [behavior]?”.

[0080] Next, the design requirements are described to the big language model in three steps. The first step is, "What functions should a [foot exercise rehabilitation device] have?" At this point, the big model outputs a series of possible functions, such as "...a series of functions, including: 1. Provide feedback on muscle strength, range of motion, and coordination, 2....., 3......, 4......", and the user selects the solution based on his or her understanding of the problem and the expected design direction. For example, the function of "providing muscle feedback" is selected, and the transformation from "requirement->function" is realized.

[0081] In the second step, the user inputs “What kind of design can demonstrate [providing muscle feedback]”, and the model outputs results such as “using sensors and / or motion tracking technology to measure muscle strength…”. After analyzing the results generated by the model, the user may think that the behavior of “using sensors to measure muscle strength” is more reliable, and the transition from “function -> behavior (export)” is achieved.

[0082] The third step is “What kind of structural design can achieve [using sensors to measure muscle strength]”, and the model answers “There are several possible structures: 1. Force sensor, 2. EMG, 3. Accelerometer, 4…”, and the transition from “behavior (derivative) -> structure” is completed.

[0083] Subsequently, the designer screens the solutions based on the relationship evaluation between behavior (derived) and behavior (expected), and forms a "function-structure" pair. For example, the designer infers that the behavior (expected) caused by the structure "force sensor" is "force sensor can be used", and the rationality of this solution is examined by evaluating the consistency between "behavior (expected)" and "behavior (derived)" (in this case, "using sensor to measure muscle strength"). For example, if the designer believes that the consistency is high from the perspective of semantic relevance, it is combined with the functional solution "providing muscle feedback" selected in this round of reasoning to form a "function-structure" mapping, "providing muscle feedback-force sensor".

[0084] After multiple rounds of dialogue with the large language model, this group finally obtained a set of "function-structure" mappings to help designers with creative ideas.

[0085] 3) Blank group G3, using the large language model to freely generate conceptual designs.

[0086] Designers in this group freely wrote prompts based on design requirements, such as "How to design a leg motion rehabilitation device?", "Design an intelligent leg motion rehabilitation device", etc. The large model may generate answers such as "...the movements they need to do in order to return to their pre-injury level...motion sensors, vibration motors, electrical muscle stimulation and other technologies. Ensure that the device is ergonomic and user-friendly."

[0087] Designers screen the design solutions generated by the large language model. If they are not satisfied with the generated results, they can order the large language model to regenerate.

[0088] All three groups required designers to draw conceptual design sketches based on the solution prompts generated by the large model and their own understanding of the design problems.

[0089] (4) Experimental evaluation

[0090] Pugh's decision index is a widely accepted concept design evaluation method, which includes three steps: selecting evaluation reference, formulating evaluation criteria and scoring. This example selects the Red Dot Award's "DANU Sports" as the evaluation reference, and refers to the concept scoring after Ulrich and Eppinger's improved Pugh evaluation, and determines novelty, feasibility, usability, functionality and cost as the evaluation criteria of this experiment, with each criterion weighted at 0.2.

[0091] This example uses a 5-point Likert scale to compare with the reference concept. The 30 concept sketches drawn by the participants were randomly mixed and compared with the reference, and their scores are shown in Table 2.

[0092] Table 2

[0093] Relative performance score Much worse than the reference concept 1 Worse than the reference concept 2 Same as reference concept 3 Better than the reference concept 4 Much better than the reference concept 5

[0094] In order to evaluate the generative effect of the method to support conceptual design ideation, the rating scores of each group of novice designers and professional designers, as well as the average score of each group, were calculated. The results are shown in Table 3.

[0095] Table 3

[0096]

[0097] C1 to C5 represent five different evaluation criteria. The scores of novices and professionals under each criterion are the weighted average of their five scores for comparison with the benchmark score. The final average score of each group is used for comparison with each other.

[0098] This embodiment uses Cronbach's alpha to evaluate the consistency between scores. Cronbach's alpha is a statistical tool that is particularly suitable for evaluating the consistency between scale scoring items or multiple raters, thereby testing their reliability. The Cronbach's alpha result for concept scoring is 0.93, indicating that the consistency between raters during the scoring process is very high.

[0099] To identify the concept score, the mean and standard deviation of each group from G1 to G3 were calculated, and the results are shown in Table 4. The average concept score of G2 The average concept score of G2 was 3.33 (SD = 0.46), which was the highest among the three groups. The average concept score of the G1 group was 2.92 (SD = 0.32), which was higher than that of the G3 group (mean 2.60, SD = 0.35). To verify whether the results were statistically significant, a one-way ANOVA was performed. The results showed that the performance of the conceptual design score on G1 and G2 was better than that on G3. The results were statistically significant (p = 0.004 < 0.01, p = 0.000 < 0.01).

[0100] Table 4

[0101] Group average value variance G1 2.92 0.32 G2 3.33 0.46 G3 2.60 0.35

[0102] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A human-machine collaborative concept design generation method based on FBS theory, characterized in that: The following steps are involved: Obtain conceptual design cases and extract keywords based on design requirements and functions, behaviors, and structures in FBS theory to construct a functional space for mapping requirements to functions, a behavioral space for mapping functions to behaviors, and a structural space for mapping behaviors to structures: Based on the logical sequence of the conceptual design process, the function space, behavior space and structure space are sequentially embedded in the large language model framework to construct a conceptual design auxiliary model, which includes a prompt word guidance module, a function solution module, a behavior solution module, a structure solution module and a design solution summary module: The prompt word guidance module obtains corresponding keywords through the prompt word template. The prompt word guidance module decomposes the design reasoning process into three sub-steps based on the function, behavior and structure in the FBS theory to construct the corresponding prompt word template; The prompt word template includes three sub-templates, which are used for function space solution, behavior space solution and structure space solution respectively. Each sub-template is a question sentence, which contains identifier specifications for capturing FBS-related knowledge in the zero-shot learning prompt setting, and placeholders for prompting users to fill in information; The sub-template for solving the function space is specifically set as follows: Identifier specification: functionality; Placeholders: design objects and requirements; That is, the user is prompted to input text content related to "design objects and requirements", and the content related to "functions" in FBS is captured in the input text content; The sub-template for behavior space solution is specifically set as follows: Identifier specification: shows; Placeholder: Function; That is, the user is prompted to enter text content related to "function" and the content related to "performance" in FBS is captured in the entered text content; The sub-template used for structure space solving has the following specific settings: Identifier specification: structure; Placeholder: behavior; That is, the user is prompted to enter text content related to "behavior", and the content related to "structure" in FBS is captured in the entered text content; The function solving module generates a corresponding function set according to the keywords of the design requirements, and the behavior solving module generates a corresponding behavior set according to the keywords of the functions in the function set; The structure solving module generates a corresponding structure set according to the keywords of the behaviors in the behavior set; The design solution summary module manually and autonomously derives the behavior corresponding to each structure based on the acquired structure set, and compares the consistency with the behavior set output by the behavior solving module. If the derived behavior does not appear in the behavior set, the structure corresponding to the structure set is eliminated, otherwise the corresponding structure is retained to obtain a structure set with correlation, and the functions corresponding to the structure set with correlation are paired to output a visual function-structure mind map; Input the design requirements into the conceptual design auxiliary model to obtain the function-structure mind map corresponding to the design requirements.

2. The method for generating human-machine collaborative concept design based on FBS theory according to claim 1 is characterized in that: The expression of the prompt word guidance module is: GC=M ptl (PT(x)) Among them, GC represents the generated text content, x represents the input keyword or phrase, PT represents the prompt template used to package the input content, and M ptl Stands for Large Language Model Framework.

3. The method for generating human-machine collaborative concept design based on FBS theory according to claim 1 is characterized in that: The large language model framework is constructed using a question-answering model and a text generation model.

4. The method for generating human-machine collaborative concept design based on FBS theory according to claim 3 is characterized in that: The hyperparameters of the large language model framework are set as follows: The temperature is set to 0.7; The maximum length of the output text is set to 256; The Top P value is set to 1; Frequency is set to 0; The presence penalty is set to 0.

5. The method for generating human-machine collaborative concept design based on FBS theory according to claim 1 is characterized in that: The function-structure mind map represents the set of implementation structures corresponding to each key function in the design solution in a visual form.

6. A human-machine collaborative concept design generation system, characterized in that: It is implemented by the human-machine collaborative concept design generation method based on FBS theory as described in any one of claims 1 to 5, including a data interaction unit, a data processing unit and a data visualization unit; The data interaction unit allows the user to input design requirements based on the provided prompt word template, screen and further improve the functions, behaviors and structures generated by the model, and add the improved results to the designed prompt word template; The data processing unit, according to the screening results of the data interaction unit, gradually infers the functions, behaviors and structures corresponding to the design requirements, and finally outputs a function-structure set that conforms to the FBS method and can break the designer's mindset in combination with user evaluation and screening; The visualization unit generates a corresponding visualization mind map according to the obtained function-structure set.

Images