Intelligent voice assistant design method based on AIGC technology

Abstract

The application discloses a kind of based on AIGC technology's intelligent voice assistant design method.Natural language form instruction of receiving user input;According to the instruction of user, determine the performance index of design product and engineering manufacturing constraint, solution, and determine structured prompt word;According to structured prompt word, by product 3D generation model, generate personalized 3D shell grid and carry out automated manufacturability analysis, after manufacturability analysis, personalized 3D shell grid and module component 3D model virtual assembly are passed, with ambient light rendering, generate high-fidelity rendering output display.The present application realizes intelligent, accurate, systematized conversion from user language to manufacturable design scheme.

Description

Technical Field

[0001] This invention relates to the field of product design technology, and in particular to a design method for an in-vehicle intelligent voice assistant based on AIGC technology. Background Technology

[0002] With the evolution of connected vehicle and smart cockpit technologies, in-vehicle voice assistants have become the core hub of human-vehicle interaction. In traditional design, voice interaction requires simulating real-world dialogue scenarios, writing dialogue scripts and interaction flows to provide a natural and logical user experience. A typical design process includes: requirements analysis (defining core functions such as navigation, music, and phone calls), development of modules such as Automatic Speech Recognition (ASR), Natural Language Understanding (NLU), Dialogue Management (DM), Natural Language Generation (NLG), and Text-to-Speech (TTS), as well as knowledge base construction and user interface design. While this process emphasizes modularity and step-by-step verification, it has significant shortcomings when dealing with large-scale, fragmented personalized needs: manually written scripts and rules struggle to cover massive user preferences, knowledge base updates are slow, and large-scale personalized deployment is difficult to complete within the product iteration cycle. Furthermore, traditional design processes heavily rely on manual definition, often lacking readily available solutions for new scenarios. Overall, traditional design methods lack the ability to efficiently transform vague and subjective user needs, making it difficult to support rapid iteration.

[0003] To address the aforementioned issues, the industry urgently needs an innovative design methodology to enhance the intelligence of in-vehicle voice assistants. Against the backdrop of the rise of AIGC (AI-Generated Content Generation) technology, the automotive industry is exploring a paradigm shift from "software-defined vehicles" to "data-defined vehicles." AIGC technology can quickly generate design solutions based on user input and iteratively optimize based on user feedback, thereby improving overall design efficiency and quality. Currently, manufacturers such as Tesla have experimented with applying AIGC to voice interaction, enabling voice assistants to have more natural conversational abilities and personalized recommendations. Companies like BMW utilize AIGC for personalized interface and content customization, improving user satisfaction.

[0004] However, there are still the following problems in how to systematically integrate AIGC technology into the engineering design process of in-vehicle voice assistants to achieve efficient and reliable digital design solutions: (1) Traditional design processes are difficult to efficiently handle massive and ambiguous personalized needs. Users’ needs for voice assistants are not limited to basic functions such as navigation and music playback, but extend to deeper experiences such as emotional interaction, multimodal perception, vehicle-home interconnection, and personalized scene adaptation. These needs are mostly in the form of natural language, emotional words or fragmented scene descriptions, which are far removed from the quantifiable performance indicators (PI) and design variables (DV) in engineering design. Designers rely on subjective experience to transform needs, which is inefficient and prone to deviation, and cannot achieve precise design that is “personalized for each person”. (2) There is a lack of systematic solutions to the inherent contradictions between cross-domain performance requirements. Intelligent in-vehicle voice assistants are multi-objective optimization systems, and their different dimensions of performance indicators (PI) often conflict with each other. For example, improving the "texture" and "luxury" of the shell may directly conflict with the goals of "lightweight" and "low cost"; enhancing the "robustness in noisy environments" of voice recognition may contradict the requirements of "small size" and "low power consumption". Traditional design methods rely on designers to make trade-offs in limited solutions, lacking the ability to actively break contradictions and seek better solutions by utilizing systematic innovation theory, resulting in insufficient design innovation and easy to fall into local optima. (3) There is an inherent contradiction between personalized needs and engineering manufacturability. Users may simultaneously require the shell of the voice assistant to be "lightweight and sturdy", "warm to the touch and scratch-resistant", and require the built-in microphone array to maintain a high recognition rate in complex noise environments. This involves multiple conflicts between material selection, structural design, acoustic structure and cost control. Traditional design methods lack systematic tools to identify and resolve these cross-domain contradictions, often leading to repeated compromises between personalization and feasibility in design solutions, limiting innovation. (4) Design knowledge is fragmented, making it difficult to form reusable and evolving corporate assets. The design of voice assistants involves multiple professional fields such as CMF, structure, acoustics, electronics, and algorithms. In the traditional process, knowledge, experience and design rules in various fields are scattered in the minds of different engineers or in scattered documents, and are not structured or digitized. This leads to difficulties in knowledge transfer, lack of unified data support for design decisions, repeated solutions to similar problems in different projects, and inability to accumulate and reuse successful innovation models. (5) Multimodal interactive hardware and software co-design is complex and iterates slowly. Modern voice assistants emphasize multimodal fusion. This means that hardware design must be deeply coupled with software interaction logic. In the traditional serial process, hardware is locked before software development. Once the software interaction logic needs to adjust the hardware layout, it will lead to expensive mold modifications. There is a lack of a platform that can perform integrated simulation and co-optimization of hardware and software solutions in the early stages of design. (6) It is difficult to establish a "generation-verification" closed loop for AI-driven design models, which leads to difficulties in implementation.Even with the introduction of AIGC model generation solutions, a key challenge remains: how to automatically and efficiently evaluate whether the generated solutions meet all performance requirements and manufacturing constraints, and how to feed the evaluation results back to the model for iterative optimization. Without an effective automatic evaluation and feedback mechanism, the AIGC generation process becomes a blind "casting a wide net," unable to achieve targeted and efficient "evolution," ultimately still requiring manual selection and modification, significantly diminishing its intelligent value. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings and defects of the existing technology and provide a design method for intelligent voice assistants based on AIGC technology. The aim is to optimize the design process of in-vehicle voice assistants by utilizing AIGC technology, and to realize the intelligent, accurate and systematic transformation from user natural language input to manufacturable 3D design solutions. This solves the problems of inefficient transformation of personalized needs, prominent contradiction between individuality and manufacturability, and AIGC-generated results being divorced from engineering practice in the traditional design process.

[0006] This invention is achieved through the following technical solution:

[0007] A design method for intelligent voice assistants based on AIGC technology, used for the intelligent design and generation of in-vehicle intelligent voice assistants, includes:

[0008] S1. Receive personalized product design instructions in natural language form input by the user on the interactive interface;

[0009] S2. Use natural language processing technology to extract keywords related to the product's performance indicators from the instructions, and determine the weight of the user's required product's performance indicators and the priority of design variables based on keyword analysis;

[0010] S3. Call the pre-built product manufacturing knowledge graph, associate the product performance index weights and design variable priorities with the concrete and abstract modules in the product manufacturing knowledge graph, and determine the product's engineering manufacturing constraints;

[0011] S4. Identify design conflicts based on the performance indicators of the concrete and abstract modules of the product, and use the TRIZ innovation library to determine solutions to the design conflicts;

[0012] S5. Determine structured prompts based on the performance indicators, engineering and manufacturing constraints, and solutions of the product's concrete and abstract modules;

[0013] S6. Call the product 3D generation model, generate a personalized 3D shell mesh based on structured prompts, and perform automated manufacturability analysis. Virtually assemble the personalized 3D shell mesh that has passed the manufacturability analysis with the 3D models of module components in the module information library, and use ambient light rendering to generate a high-fidelity rendering for output display.

[0014] Preferably, after step S6, the method further includes:

[0015] S7. Receive the user's feedback optimization instructions via natural voice input, return to step S2, and after at least one round of optimization processing, output the final 3D product rendering.

[0016] Preferably, the performance index weights and design variable priorities for user-demanded products are determined based on keyword analysis, including the following steps:

[0017] The weights of different performance metrics are determined based on the term frequency-inverse document frequency and / or user emphasis of the keywords, forming a user evaluation diagonal matrix, which is composed of the weights of multiple performance metrics.

[0018] The comprehensive evaluation language matrix is ​​calculated based on the user evaluation diagonal matrix and the manufacturer's language evaluation matrix. The weighted importance score of each design variable is obtained, and the priority of the design variables is determined based on the weighted importance score of the design variables.

[0019] In this context, each element of the manufacturer's language evaluation matrix represents the influence strength value of different design variables on different performance indicators; the comprehensive evaluation language matrix represents the manufacturer's expertise and user's personalized preferences, and its column vectors reflect the weighted importance scores of each design variable under the current user preferences.

[0020] Preferably, the concrete modules and abstract modules of the product are pre-built and stored in the product module digital carrier information database, and are defined and distinguished based on the system composition and hardware composition of the intelligent vehicle voice assistant. The concrete module includes at least a shell, screen, audio system, physical buttons, status indicator lights, microphone array board, motherboard, DMS camera cover, and sound outlet mesh cover; the abstract module includes at least a main acoustic cavity, air duct heat dissipation structure, microphone anti-vibration suspension structure, and light guide column.

[0021] Preferably, the product manufacturing knowledge graph uses a database with at least module nodes, performance index nodes, design variable nodes, material nodes, and processing technology nodes of each concrete and abstract module as entities, and is constructed by the relationships between each node; the relationships between nodes at least include relationships based on material composition, influence, applicability, and influence intensity; material nodes include attribute key-value pairs that record material design variables.

[0022] Preferably, the TRIZ innovation library includes a TRIZ contradiction matrix formed by mapping identified design contradictions to 39 general engineering parameters of TRIZ, and solutions to resolve design contradictions associated with the inventive principles in the TRIZ contradiction matrix; the design contradictions are determined by analyzing the relationships between product performance indicators; and the solutions are obtained by searching relevant databases.

[0023] Preferably, the structured cue words include at least geometric constraints, material constraints, process constraints, style constraints, cost constraints, and TRIZ innovation principle constraints.

[0024] Preferably, the structured prompts are based on an explicit constraint embedding strategy, directly describing the manufacturing constraints that the product design must meet in the natural language prompts.

[0025] Preferably, the manufacturability analysis includes at least an analysis of whether the product’s minimum wall thickness, sharp corners, and overhang structure exceed the limitations of the selected process.

[0026] Preferably, the design variables include at least thermal conductivity, flexural strength, tensile strength, impact strength, corrosion resistance, Shore hardness, density, cost, dimensions, processing accuracy, style, emotional appeal, UV resistance, low-temperature brittleness, heat distortion temperature, coefficient of thermal expansion, insulation strength, electrical conductivity, damping characteristics, flame retardancy, and light transmittance; the performance indicators include at least three categories: appearance, function, and economy.

[0027] This invention uses Language Evaluation Matrix (MLEM / CLEM) and natural language processing technology to systematically map users' subjective and vague language descriptions into quantifiable engineering performance indicators (PI) and design variables (DV), solving the problems of traditional design relying on experience and being inefficient, and realizing the efficient and accurate transformation of large-scale personalized needs.

[0028] This invention provides a systematic and innovative solution to the performance conflicts that inevitably arise in design by constructing an innovation library that integrates TRIZ theory and technical knowledge and combining it with manufacturing knowledge graphs for resource matching. It breaks away from the simple trade-off between individuality and feasibility in traditional design and systematically resolves the core contradiction between individuality and manufacturability.

[0029] This invention fundamentally guides AIGC to adhere to manufacturing boundaries during creative brainstorming by embedding engineering constraints (such as materials, processes, and costs) and innovation guidance into the prompts of the generated model. This ensures the engineering feasibility of the AIGC-generated solutions, and the generated 3D solutions have high feasibility, thus realizing a closed loop between "creativity" and "manufacturing".

[0030] This invention can automatically generate diverse design schemes that differ in shape, CMF (color, material, process), and internal structural layout, each meeting the optimal performance combination and manufacturing requirements, based on different user preferences. This improves the design quality and innovation of intelligent in-vehicle voice assistants, and greatly enhances the market compatibility and competitiveness of the products. Attached Figure Description

[0031] Figure 1This is a flowchart illustrating the design method of an intelligent voice assistant based on AIGC technology according to the present invention. Detailed Implementation

[0032] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0033] See Figure 1 As shown in the exemplary embodiment of this application, the intelligent voice assistant design method based on AIGC technology is used for the intelligent design and generation of in-vehicle intelligent voice assistants, including:

[0034] S1. Receive personalized product design instructions in natural language form input by the user on the interactive interface;

[0035] S2. Use natural language processing technology to extract keywords related to the product's performance indicators from the instructions, and determine the weight of the user's required product performance indicators and the priority of design variables based on keyword analysis;

[0036] S3. Call the product manufacturing knowledge graph, associate the weights of product performance indicators and the priority of design variables with the concrete and abstract modules in the product manufacturing knowledge graph, and determine the engineering manufacturing constraints of the product.

[0037] S4. Identify design conflicts based on the performance indicators of the concrete and abstract modules of the product, and use the TRIZ innovation library to determine solutions to the design conflicts;

[0038] S5. Determine structured prompts based on the performance indicators, engineering and manufacturing constraints, and solutions of the product's concrete and abstract modules;

[0039] S6. Call the product 3D generation model, generate a personalized 3D shell mesh based on structured prompts, and perform automated manufacturability analysis. Virtually assemble the personalized 3D shell mesh that has passed the manufacturability analysis with the 3D models of the module components called in the product manufacturing knowledge graph, and use ambient light rendering to generate a high-fidelity rendering output.

[0040] The instructions for personalized product design in natural language form described in this application can be natural speech or text.

[0041] In some embodiments, after step S6, the following steps are also included:

[0042] S7. Receive the user's feedback optimization instructions via natural voice input, return to step S2, and after at least one round of optimization processing, output the final 3D product rendering.

[0043] In this application, after step S6, based on the user's feedback on the output display of the high-fidelity effect image, the optimization instructions are input into the model through natural language. The model then optimizes the product design according to the feedback optimization instructions, executing steps S2-S6. Through multiple rounds of feedback optimization, the final satisfactory 3D design drawing of the product can be generated.

[0044] In one alternative embodiment, determining the performance index weights and design variable priorities of user-demanded products based on keyword analysis includes the following steps:

[0045] The weights of different performance metrics are determined based on the term frequency-inverse document frequency and / or user emphasis of the keywords, forming a user evaluation diagonal matrix, which is composed of the weights of multiple performance metrics.

[0046] The comprehensive evaluation language matrix is ​​calculated based on the user evaluation diagonal matrix and the manufacturer's language evaluation matrix. The weighted importance score of each design variable is obtained, and the priority of the design variables is determined based on the weighted importance score of the design variables.

[0047] In this context, each element of the manufacturer's language evaluation matrix represents the influence strength value of different design variables on different performance indicators; the comprehensive evaluation language matrix represents the manufacturer's expertise and user's personalized preferences, and its column vectors reflect the weighted importance scores of each design variable under the current user preferences.

[0048] Specifically, the design method in this application is based on an AI-powered intelligent design system. When in use, the system receives descriptions input by the user in text, voice, or other forms, such as, "I want a voice assistant that looks high-tech, has a premium feel, is comfortable to the touch, doesn't easily leave fingerprints, and can clearly hear my commands while driving." The system uses NLP techniques (such as keyword extraction and sentiment / intensity analysis) to analyze and identify words related to performance indicators (PIs), such as "high-tech" → "design" and "feel"; "feel comfortable to the touch" → "feel"; "doesn't easily leave fingerprints" → "easy to clean and maintain"; "clearly hear commands" → associated with acoustic performance, such as "high-end," with "feel," "clear," and "doesn't easily leave fingerprints" being strong, and "comfortable" being medium. The system assigns weights to the identified PIs based on word frequency-inverse document frequency or user emphasis. For example, "shape" weight 0.25, "color" weight 0.15, "texture" weight 0.30, "tactile feel" weight 0.10, "abrasion resistance" weight 0.10, and "acoustic performance" weight 0.10; finally, a User Evaluation Diagonal Matrix (UEDM) is formed: Calculate the Comprehensive Language Evaluation Matrix (CLEM);

[0049] ;

[0050] This comprehensive language evaluation matrix integrates the manufacturer's expertise and the user's personalized preferences. Its column vectors reflect the weighted importance scores of each design variable (DV) under the current user preferences. For example, the calculation results might show that for this user, the weighted scores for "Shore Hardness" (affecting texture and scratch resistance) and "Damping Characteristics" (affecting the internal acoustic environment of the voice assistant) are significantly higher than others. This indicates key optimization directions for subsequent design. This is a manufacturer's language evaluation matrix used to construct the mapping relationship between user requirements and engineering parameters, bridging the gap between user language and engineering parameters.

[0051] The manufacturer language evaluation matrix is ​​constructed in the following manner;

[0052] For example, a focus group composed of structural engineers, acoustic engineers, CMF designers, and manufacturing process experts can identify key design variables (DV) and performance indicators (PI) for intelligent in-vehicle voice assistants. Specifically, the group, through discussion, uses linguistic variables, as shown in Table 1, to evaluate the strength of the impact of each design variable on each performance indicator (e.g., the impact of flexural strength on the "support" performance indicator is "VH"). Subsequently, triangular fuzzy numbers are used to convert the linguistic variables into numerical values, forming the Manufacturer's Linguistic Evaluation Matrix (MLEM).

[0053] ;

[0054] Where m is the number of design variables and q is the number of performance indicators.

[0055] The manufacturer language evaluation matrix is ​​constructed through the following steps:

[0056] Construct a triangular fuzzy evaluation matrix:

[0057] ;

[0058] The evaluation of the i-th design variable and the j-th performance index by the r-th expert is represented by a triangular fuzzy number:

[0059] The three parameters represent the corresponding lower bound, median, and upper bound, respectively.

[0060] Aggregating the evaluation results of k experts yields a fuzzy evaluation matrix for the manufacturer:

[0061] ;

[0062] in:

[0063]

[0064] Furthermore, the manufacturer fuzzy evaluation matrix is ​​defuzzified, and the manufacturer language evaluation matrix is ​​obtained using the centroid method:

[0065] ;

[0066] in:

[0067] .

[0068] Table 1. Strength of Triangular Fuzzy Number Relationships

[0069]

[0070] In some embodiments, the design variables include at least thermal conductivity, flexural strength, tensile strength, impact strength, corrosion resistance, Shore hardness, density, cost, dimensions, processing accuracy, style, emotional appeal, UV resistance, low-temperature brittleness, heat distortion temperature, coefficient of thermal expansion, insulation strength, electrical conductivity, damping characteristics, flame retardancy, and light transmittance; the performance indicators include at least three categories: appearance, function, and economy.

[0071] The Manufacturer Language Evaluation Matrix (MLEM) constructed in this embodiment is shown in Table 2.

[0072] Table 2 Manufacturer Language Evaluation Matrix

[0073]

[0074] In this application, a computable and callable knowledge base is constructed for the design process. A digital carrier of product modules and a manufacturing knowledge graph are pre-built and stored in the system. The digital carrier of product modules is constructed in the following way: First, a 3D generative model is constructed, such as using Get3D based on the GAN architecture. Blender is used to standardize the product model into a boundary cube. Then, the same standard lighting settings and camera poses randomly sampled in the upper hemisphere are configured through a script. An RGBAD image is exported through Blender's built-in rendering engine. The Get3D model is fine-tuned using the RGBAD image to obtain the required 3D generative model.

[0075] Furthermore, based on the system and hardware configuration of the intelligent in-vehicle voice assistant, different modules of the product are defined, and relevant design variables are associated with each module to construct a module information database and manufacturing knowledge graph. Modules can be divided into concrete modules and abstract modules. Concrete modules (with a defined form) include the outer shell, screen, audio system, physical buttons, status indicator lights, microphone array board, motherboard, DMS camera cover, and sound outlet grille. Abstract modules (functionally defined, form to be generated) include the main acoustic cavity, air duct heat dissipation structure, microphone anti-vibration suspension structure, and light guide column.

[0076] Each module is associated with relevant design variables (DV) and performance indicators (PI). Taking the "shell module" as an example, its associated design variables (DV) include different material properties such as the flexural strength, cost, and corrosion resistance of ABS plastic, or the flexural strength, light transmittance, and cost of photosensitive resin 9400A. Performance indicators are used to characterize user perception or system performance requirements, including but not limited to appearance features and surface treatment effects. Among them, "shape: streamlined" is used to describe the perceived effect of the product's geometry; "surface treatment: matte" is used to describe the visual and tactile performance of the product's surface. The above-mentioned performance indicators (PI) as perceived attributes all participate in the subsequent evaluation process.

[0077] In this application, the product manufacturing knowledge graph uses a database with at least module nodes, performance index nodes, design variable nodes, material nodes, and processing technology nodes of each concrete and abstract module as entities, and is constructed by the relationships between each node; the relationships between nodes include at least the relationships of being made of materials, influencing, applicable to, and influencing intensity; the material nodes include attribute key-value pairs that record material design variables.

[0078] Taking "Shell" as an example, a node "Shell" is created, which is connected to the "ABS" node through the "Made of Material" relationship. The "ABS" node has attribute key-value pairs {"Flexural Strength": 97, "Cost": 5, ...}. Simultaneously, the "Shell" node is connected to the PI node "Texture" through the "Influence" relationship, and "Texture" has an "Influence Strength" relationship with the DV node "Shore Hardness" inherited from the Manufacturer Language Evaluation Matrix (MLEM). The processing technology ("Processing Technology" worksheet) is also added as a node, connected to the material node through the "Applies To" relationship (e.g., "ABS" applies to "Injection Molding").

[0079] In this embodiment of the application, in order to achieve the integration and conflict resolution of manufacturer knowledge and user preferences, a "rehearsal" and "decision-making" of the design scheme generation are also performed before generating structured prompt words.

[0080] That is, after mapping user preferences to performance indicators (PIs) and design variables (DVs), an engineering constraint space is further constructed for the generation of structured prompts. Based on the product manufacturing knowledge graph, which encompasses static knowledge such as modules, PIs, DVs, materials, and processes, the product manufacturing knowledge graph is instantiated and constraint extracted in a product task-oriented manner to obtain the engineering manufacturing constraints for product instantiation, including:

[0081] Starting with DV (Design Documentation), the system extracts manufacturing knowledge entities (such as optional materials, applicable processes, and tolerance capabilities) related to the current user's needs from design specifications, process documents, and equipment capability libraries, forming instantiated material-process combinations. For example, when a user prefers "high-quality feel" and "scratch-resistant," the system retrieves the "Shore Hardness" and "Surface Treatment" nodes from the knowledge graph and instantiates them as "PC+ABS+Matte UV Coating."

[0082] By integrating the performance index (PI) weighted vector of user preference transformation with instantiated manufacturing knowledge, a comprehensive language evaluation matrix (CLEM) is generated. This matrix quantifies the importance score of each design variable (DV) under user preferences, providing a unified semantic input for design variable prioritization, contradiction identification, and prompt word generation.

[0083] In this embodiment, to ensure that the AI-generated design meets the standards of high-quality production, it is necessary to integrate the demand and supply sides, transforming user needs into precise prompts and effectively resolving potential design contradictions. Against this backdrop, an Innovative Design Method (IDM) based on the Theory of Inventive Problem Solving (TRIZ) is creatively proposed. This method constructs a TRIZ innovation library by systematically modeling design contradictions, assisting generative AI in achieving effective innovation in complex design tasks. The TRIZ innovation library of this application includes a TRIZ contradiction matrix formed by mapping identified design contradictions to 39 general engineering parameters of TRIZ, and solutions to resolve design contradictions associated with the inventive principles in the TRIZ contradiction matrix. The design contradictions are determined by analyzing the relationships between product performance indicators; the solutions are obtained by searching relevant databases.

[0084] Specifically, this can be achieved by analyzing the relationships between various performance indicators (PIs) in the Comprehensive Language Evaluation Matrix (CLEM) or by using knowledge graph reasoning or expert rules. For example, analyzing performance indicator conflicts in the CLEM reveals that improving "lightweighting" often leads to the use of lower-strength materials, thus weakening "support," which constitutes a design contradiction. Similarly, using knowledge graph reasoning or expert rules, it is found that improving texture may increase cost, also constituting a design contradiction. These identified design contradictions are mapped to the 39 general engineering parameters of TRIZ, such as "weight of moving objects" and "strength," forming a TRIZ contradiction matrix. Querying the TRIZ contradiction matrix yields recommended inventive principles, such as the principle of segmentation and the principle of composite materials. To further obtain specific solutions in this field (consumer electronics, automotive devices), text analysis tools such as AntConc are used to search relevant databases, such as patent databases, using keywords such as "lightweighting," "high strength," "shell," and "acoustic structure." By using text mining and solution extraction techniques, specific technical means to solve similar design contradictions are extracted from the database, such as "a lightweight and high-strength structure using a magnesium alloy skeleton and composite plastic coating" and "a design that uses internal reinforcing ribs to optimize topology and reduce weight while improving stiffness". The extracted specific solutions are associated with the aforementioned TRIZ invention principles and stored in the TRIZ innovation library in a structured manner, thus realizing the integration of domain knowledge.

[0085] In the conflict between lightweight and high strength, the corresponding TRIZ parameters are weight vs. strength, the solution is topology optimization with stiffeners, and the term is lightweight design.

[0086] To further obtain solutions to design contradictions in this field, a dedicated corpus was constructed.

[0087] Table 3 Construction of the TRIZ Innovative Corpus for the Intelligent Voice Assistant

[0088]

[0089] Table 4 TRIZ Innovative Terminology Database for Intelligent Voice Assistants

[0090]

[0091] Specifically, when designing a product for the system, solutions can be proposed based on historically stored solutions, or relevant databases can be queried in real time to update the data, and solutions can be proposed based on the real-time queried data.

[0092] Incorporating manufacturability constraints into prompt engineering is a crucial step in the successful implementation of AIGC-driven design systems. Large language models (such as GPT-4, Claude, and Gemini) or multimodal generative models (such as Stable Diffusion and DALLE) are essentially language-driven content generation systems, lacking awareness of engineering constraints. Therefore, the design needs to leverage prompts that consider manufacturability to guide the model in incorporating manufacturing limitations during the generation process, ensuring that the output is not only creative but also engineering-feasible.

[0093] In some embodiments, the structured prompts are based on an explicit constraint embedding strategy, directly describing the manufacturing constraints that the product design must meet in the natural language prompts.

[0094] The structured prompts output by the system in this application include at least geometric constraints, material constraints, process constraints, style constraints, cost constraints, and TRIZ innovation principle constraints: Geometric constraints: To prevent the generation of structures that exceed manufacturing limits, the key dimensional limitations of the product should be clearly indicated in the prompts. For example: "Please design a connector shell with an outer diameter not exceeding 100mm and a wall thickness not less than 3mm for injection molding." By setting constraints such as "maximum size," "thickness limit," and "radius of curvature," the system guides the generation of geometric features that conform to the processing dimensions, preventing excessively thin, thin, or highly free-form surface structures.

[0095] Material property constraints: For the previously determined optimal manufacturing elements, different materials have different limitations and considerations. Therefore, it is necessary to extract relevant manufacturing knowledge from the manufacturing knowledge graph and limit or suggest the range of materials to be used in the prompts. For example, "Use 3D printable polyamide materials (such as PA12), while taking into account wear resistance and UV resistance." or "The design must consider the use of aerospace-grade aluminum alloy 7075, which can support T6 heat treatment." The material description not only limits the physical properties but also indirectly constrains the available processing methods.

[0096] Process adaptability: Similar to materials, the practicality of a design directly depends on whether it can be perfectly implemented through the target process. The prompts should reflect its compatibility with the manufacturing process, for example: "The structure must be suitable for CNC milling, avoiding dead corners in the internal cavity and multi-axis interference areas." "Please prioritize injection molding demolding designs, avoiding the use of negative draft angles and complex undercut structures." "Suitable for mold opening with a double parting surface design, minimizing the use of sliders."

[0097] Design Style Constraints: To ensure the generated results possess a unified aesthetic orientation while meeting functional and manufacturing requirements, the prompts often need to explicitly specify the style preference of the target design. Design style not only influences the product's formal language (such as lines, proportions, and textures) but also indirectly guides the generated model's preferences in shape, color scheme, and composition. For example: "The design should embody a Nordic minimalist style, emphasizing geometric symmetry, undecorated edges, and primarily using neutral gray and white colors." or "Adopt a futuristic industrial style, with sharp outlines and exposed structures, referencing cyberpunk aesthetics." Style prompts serve as high-level semantic constraints, influencing the form and language direction early in model generation, thereby avoiding the generation of solutions that contradict brand tone or user expectations.

[0098] In this way, the prompts not only convey the user's explicit preferences, but also introduce the tacit knowledge from the manufacturing end, enabling the generative model to consider both design logic and engineering constraints when understanding and generating.

[0099] Meanwhile, during the personalized design process, users often raise conflicting requirements such as needing both "lightweight" and "high-intensity," or "low-cost" and "functionally complex." Traditional prompts struggle to accurately express this complexity. By introducing TRIZ's contradiction identification logic and innovative solution principles, these design contradictions can be made explicit and embedded as systematic language fragments into the prompt. This allows the generative model to proactively consider conflict coordination and innovative balance during content generation, thereby generating solutions that satisfy manufacturing constraints while possessing greater design logic and creative value.

[0100] This application's innovative approach achieves a logical closed loop of "from demand → contradiction → prompt → design," a crucial step in effectively transferring engineering innovation knowledge into the AIGC model. It also provides a clear semantic interface and problem-oriented approach for manufacturability design generation assisted by a large model, explicitly injecting the designer's innovative logic into the input context of the generative model. By combining a three-layered prompt structure of user preferences—engineering constraints—TRIZ innovation direction, the generation system not only avoids common "mechanical repetitive design" but also enables diverse and creative design exploration while preserving functional feasibility. It not only satisfies "manufacturability" but also encourages "creative manufacturing," enabling the generative model to systematically overcome design contradictions while adhering to manufacturing constraints.

[0101] Structured language templates are more effective than free expression in guiding models to produce compliant designs. For example:

[0102] Table 5. Prompt Attributes and Examples

[0103]

[0104] Specifically, in step S6, after generating the 3D style of the product from the generated model, an automated manufacturability analysis (such as using software like Netfabb) is performed on the generated 3D mesh to check whether the minimum wall thickness, sharp corners, and overhang structures exceed the limits of the selected process (SLA / FDM).

[0105] Once approved, the personalized shell mesh is virtually assembled with standard 3D models of screen modules, button components, etc., retrieved from the product manufacturing knowledge graph. High-fidelity rendering is then performed using unified HDR ambient lighting in KeyShot or Blender.

[0106] Present the renderings to the user. If the user feedback is that the design lacks a sense of technology or that they "want more warmth," the designer can translate this feedback into adjustments to the performance metrics (PI weights) of the "shape" or "color." After updating the UEDM, start a new optimization cycle from step 2 until the user confirms.

[0107] Through the above steps, this invention achieves a complete and intelligent transformation from a vague user description to a highly personalized, engineering-detailed, and manufacturable 3D design solution that can be directly used for subsequent development.

[0108] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention.

[0109] Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of the equivalents of the claims be included within the invention.

[0110] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A design method for an intelligent voice assistant based on AIGC technology, characterized in that, Intelligent design and generation for in-vehicle intelligent voice assistants, including: S1. Receive personalized product design instructions in natural language form input by the user on the interactive interface; S2. Use natural language processing technology to extract keywords related to the product's performance indicators from the instructions, and determine the weight of the user's required product performance indicators and the priority of design variables based on keyword analysis; S3. Call the product manufacturing knowledge graph, associate the weights of product performance indicators and the priority of design variables with the concrete and abstract modules in the product manufacturing knowledge graph, and determine the engineering manufacturing constraints of the product. S4. Identify design conflicts based on the performance indicators of the concrete and abstract modules of the product, and use the TRIZ innovation library to determine solutions to the design conflicts; S5. Determine structured prompts based on the performance indicators, engineering and manufacturing constraints, and solutions of the product's concrete and abstract modules; S6. Call the product 3D generation model, generate a personalized 3D shell mesh based on structured prompts, and perform automated manufacturability analysis. Virtually assemble the personalized 3D shell mesh that has passed the manufacturability analysis with the 3D models of module components in the module information library, and use ambient light rendering to generate a high-fidelity rendering for output display.

2. The intelligent voice assistant design method based on AIGC technology according to claim 1, characterized in that, Following step S6, the following is also included: S7. Receive the user's feedback optimization instructions via natural voice input, return to step S2, and after at least one round of optimization processing, output the final 3D product rendering.

3. The intelligent voice assistant design method based on AIGC technology according to claim 1, characterized in that, Based on keyword analysis, determine the performance indicator weights and design variable priorities for user-demanded products, including the following steps: The weights of different performance metrics are determined based on the term frequency-inverse document frequency and / or user emphasis of the keywords, forming a user evaluation diagonal matrix, which is composed of the weights of multiple performance metrics. The comprehensive evaluation language matrix is ​​calculated based on the user evaluation diagonal matrix and the manufacturer's language evaluation matrix. The weighted importance score of each design variable is obtained, and the priority of the design variables is determined based on the weighted importance score of the design variables. In this context, each element of the manufacturer's language evaluation matrix represents the influence strength value of different design variables on different performance indicators; the comprehensive evaluation language matrix represents the manufacturer's expertise and user's personalized preferences, and its column vectors reflect the weighted importance scores of each design variable under the current user preferences.

4. The intelligent voice assistant design method based on AIGC technology according to claim 1, characterized in that, The concrete and abstract modules of the product are pre-built and stored in the product module digital carrier information database. They are defined and distinguished based on the system composition and hardware composition of the intelligent vehicle voice assistant. The concrete module includes at least a shell, screen, audio system, physical buttons, status indicator lights, microphone array board, motherboard, DMS camera cover, and sound outlet mesh cover; the abstract module includes at least a main acoustic cavity, air duct heat dissipation structure, microphone anti-vibration suspension structure, and light guide column.

5. The intelligent voice assistant design method based on AIGC technology according to claim 1, characterized in that, The product manufacturing knowledge graph uses a database with at least module nodes, performance index nodes, design variable nodes, material nodes, and processing technology nodes of each concrete and abstract module as entities, and is constructed by the relationships between each node; the relationships between nodes include at least the relationships of being made of materials, influencing, applicable to, and influencing intensity; material nodes include attribute key-value pairs that record material design variables.

6. The intelligent voice assistant design method based on AIGC technology according to claim 1, characterized in that, The TRIZ innovation library includes a TRIZ contradiction matrix formed by mapping identified design contradictions to 39 general engineering parameters of TRIZ, and solutions to resolve design contradictions associated with the inventive principles in the TRIZ contradiction matrix; the design contradictions are determined by analyzing the relationships between product performance indicators; the solutions are obtained by searching relevant databases.

7. The intelligent voice assistant design method based on AIGC technology according to claim 1, characterized in that, The structured cue words include at least geometric constraints, material constraints, process constraints, style constraints, cost constraints, and TRIZ innovation principle constraints.

8. The intelligent voice assistant design method based on AIGC technology according to claim 1, characterized in that, The structured prompts are based on an explicit constraint embedding strategy, directly describing the manufacturing constraints that the product design must meet in the natural language prompts.

9. The intelligent voice assistant design method based on AIGC technology according to claim 1, characterized in that, The manufacturability analysis includes at least the analysis of whether the product’s minimum wall thickness, sharp corners, and overhang structures exceed the limitations of the selected process.

10. The intelligent voice assistant design method based on AIGC technology according to claim 1, characterized in that, The design variables include at least thermal conductivity, flexural strength, tensile strength, impact strength, corrosion resistance, Shore hardness, density, cost, dimensions, processing accuracy, style, emotional vocabulary, UV resistance, low-temperature brittleness, heat distortion temperature, coefficient of thermal expansion, insulation strength, electrical conductivity, damping characteristics, flame retardancy, and light transmittance; the performance indicators include at least three categories: appearance, function, and economy.

Images