CN118839606A - Product intelligent design method and system based on domain knowledge enhanced large language model - Google Patents

Abstract

The present invention discloses a product intelligent design method and system based on a domain knowledge enhanced large language model, including: step S1, obtaining a design task data set in the product design field; step S2, obtaining a large language model based on multi-task adaptation and knowledge enhancement according to the design task data set and product design domain knowledge; step S3, obtaining a product design generation agent through a product design task tool set according to the large language model based on multi-task adaptation and knowledge enhancement; step S4, the product design generation agent adjusts the product design plan in real time according to user feedback. The technical solution of the present invention solves the problems of fragmentation, lack of systematicness and automation in the existing product design process.

Description

Product intelligent design method and system based on domain knowledge enhanced large language model

Technical Field

The present invention belongs to the technical field of product design, and in particular relates to a product intelligent design method and system based on domain knowledge enhanced large language model.

Background Art

The existing product design process is a linear, function-oriented approach that suffers from fragmentation, lack of systematicity and automation, and is difficult to adapt to the challenges brought about by intensified market competition and diversified user needs.

Summary of the invention

The technical problem to be solved by the present invention is to provide a product intelligent design method and system based on domain knowledge enhanced large language model, which solves the problems of fragmentation, lack of systematicness and automation in the existing product design process.

To achieve the above object, the present invention adopts the following technical solution:

A product intelligent design method based on domain knowledge-enhanced large language model, comprising:

Step S1, obtaining a design task dataset in the product design field;

Step S2: obtaining a large language model based on multi-task adaptation and knowledge enhancement according to the design task dataset and product design domain knowledge;

Step S3, obtaining a product design generation agent through a product design task tool set according to a large language model based on multi-task adaptation and knowledge enhancement;

Step S4: The product design generation agent adjusts the product design plan in real time according to user feedback.

Preferably, the design task data set includes: design drawings, product specifications, market analysis reports, consumer feedback and comments, material databases, and manufacturing and process parameters; product design domain knowledge includes: knowledge graphs and expert knowledge; the product design task toolset includes: competitive product analysis, design drawing generation, professional knowledge retrieval, and design solution generation.

Preferably, step S2 comprises:

According to the design task dataset and product design domain knowledge, train a multi-design task adapter based on the Transformer model;

According to the trained Transformer-based multi-design task adapter and product design domain knowledge, a large language model based on multi-task adaptation and knowledge enhancement is constructed based on semantic retrieval.

Preferably, the product is a range hood.

The present invention also provides a product intelligent design system based on domain knowledge enhanced large language model, comprising:

An acquisition device, used for acquiring a design task data set in the field of product design;

A training device, used for training a large language model based on multi-task adaptation and knowledge enhancement according to a design task dataset and product design domain knowledge;

A generating device, for obtaining a product design generating agent through a product design task tool set according to a trained large language model based on multi-task adaptation and knowledge enhancement;

The design device is used to adjust the product design scheme in real time according to user feedback through the product design generation agent.

Preferably, the design task data set includes: design drawings, product specifications, market analysis reports, consumer feedback and comments, material databases, and manufacturing and process parameters; product design domain knowledge includes: knowledge graphs and expert knowledge; the product design task toolset includes: competitive product analysis, design drawing generation, professional knowledge retrieval, and design solution generation.

Preferably, the processing device comprises:

A training unit, used to train a multi-design task adapter based on the Transformer model according to the design task dataset and product design domain knowledge;

A construction unit is used to build a large language model based on multi-task adaptation and knowledge enhancement based on semantic retrieval according to the trained Transformer-based multi-design task adapter and product design domain knowledge.

Preferably, the product is a range hood.

The present invention obtains a large language model based on multi-task adaptation and knowledge enhancement according to a design task data set, product design domain knowledge and a product design task toolset, and obtains a product design generation agent; the product design generation agent adjusts the product design plan in real time according to user feedback; the present invention can overcome the shortcomings of existing linear and function-oriented design methods, and expand to a wider range of demand research, user participation, market analysis, innovative design and user experience, thereby realizing the automation, intelligence and systematization of the product design process.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention 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 embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

FIG1 is a flow chart of a method for intelligent product design based on domain knowledge-enhanced large language model according to an embodiment of the present invention;

FIG2 is a schematic diagram of the working principle of the product intelligent design method based on domain knowledge enhanced large language model according to an embodiment of the present invention;

FIG3 is a schematic diagram of the structure of a multi-design task adapter;

FIG4 is a schematic diagram of training a multi-design task adapter based on a Transformer model;

FIG5 is a schematic diagram of building a large language model based on multi-task adaptation and knowledge enhancement.

DETAILED DESCRIPTION

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

Embodiment 1:

As shown in FIGS. 1 and 2 , an embodiment of the present invention provides a product intelligent design method based on domain knowledge-enhanced large language model, including:

Step S1, obtaining a design task dataset in the product design field;

Step S2: obtaining a large language model based on multi-task adaptation and knowledge enhancement according to the design task dataset and product design domain knowledge;

Step S3, obtaining a product design generation agent through a product design task tool set according to a large language model based on multi-task adaptation and knowledge enhancement;

Step S4: The product design generation agent adjusts the product design plan in real time according to user feedback.

As an implementation manner of an embodiment of the present invention, the product is a range hood; the design task data set includes: design drawings, product specifications, market analysis reports, consumer feedback and comments, material databases, and manufacturing and process parameters; the product design domain knowledge includes: knowledge graphs and expert knowledge.

Further, design drawings: design drawings of the range hood, such as 2D and 3D CAD models; range hood annotation information, such as component names, sizes, and materials.

Product Specifications: Product specifications in documented form, such as power, dimensions, materials used, wind and noise levels; product manuals and assembly guides.

Market analysis report: analysis of competitive products, user orders, user preferences and market trends.

Consumer feedback and reviews: User reviews and ratings from e-commerce platforms and surveys; user sentiment, preferences, and product improvement points.

Materials database: physical and chemical properties of various materials, such as heat resistance, strength, color and texture; material suppliers and their product catalog data.

Manufacturing and process parameters: Parameters related to the manufacturing process, such as machine type, processing time, and energy consumption.

As an implementation of the embodiment of the present invention, step S2 includes:

According to the design task dataset and product design domain knowledge, train a multi-design task adapter based on the Transformer model;

According to the trained Transformer-based multi-design task adapter and product design domain knowledge, a large language model based on multi-task adaptation and knowledge enhancement is constructed based on semantic retrieval.

Further, as shown in FIGS. 3 and 4 , the multi-design task adapter includes:

1. Multimodal design task dataset and product design domain knowledge processing:

1) Multimodal design task dataset: Text data processing: Use tools such as jieba and THULAC for word segmentation, use HanLP for part-of-speech tagging and named entity recognition (NER), extract keywords and phrases through TF-IDF and TextRank algorithms, and use Stanford CoreNLP for dependency syntax analysis. Image data processing: Use ViT (Visual Transformer) self-attention mechanism to identify and locate target objects in images, and achieve accurate pixel-level classification through semantic segmentation tasks. In terms of feature extraction, ViT encodes the deep features of the image layer by layer to generate rich and discriminative vector representations. In addition, combined with text generation technology, ViT can generate natural language that accurately describes the content of the image, bridging the gap between visual information and language expression. Structured data processing: For structured data such as Excel tables, first perform data cleaning (remove null values, formatting and other preprocessing operations), then extract specific fields and their corresponding values as entities and relationships, and convert them into common formats (such as RDF, JSON-LD).

2) Product design domain knowledge: Knowledge graph construction: Using Neo4j tools, triples (entity-relationship-entity) are extracted from the processed data to construct a graph containing multi-perspective knowledge such as style and space. The knowledge graph provides systematic knowledge in the field of product design, which can be used for subsequent model training and knowledge enhancement.

2. Task-Knowledge (Prompt Template) Module:

1) Prompt template design and instance generation:

Design a prompt template: Based on the design task dataset and knowledge graph, analyze the task objectives and design a prompt template containing placeholders such as [mask].

Generate Prompt instances: Traverse the knowledge graph and automatically generate a large number of Prompt instances based on the designed Prompt template. For the [mask] position, find and fill the corresponding entity from the knowledge graph; for [image features], extract the corresponding features from the image data and fill them in the placeholder. Finally, generate a large number of Prompt instances that integrate task semantic descriptions, domain knowledge references, and multimodal data features, provide high-quality training data, and ensure that the model can simultaneously learn semantic understanding, knowledge fusion, and multimodal representation.

3. Knowledge injection and multi-task adaptation module:

Task semantic vector encoding: Use a pre-trained language model (such as BERT) to encode the part of the prompt that describes the task semantics, generate a task semantic vector, and input it into the multi-design task adapter 1 (task semantic adaptation module) shown in Figure 4.

Domain knowledge vector encoding: Use knowledge graph embedding technology (such as TransE, RotatE, etc.) to encode entities and relationships in the knowledge graph, fuse them to obtain domain knowledge vector representation, and input it into the multi-design task adapter 2 (knowledge adaptation module) shown in Figure 4.

Multimodal feature vector encoding: For multimodal data such as [image features] and [design description] in Prompt, use visual transformer (such as ViT) and language model (such as BERT) to encode them respectively, and concatenate or fuse the encoded multimodal feature vectors and inject them into the multi-design task adapter as input. The outputs of multiple multi-design task adapters are combined as features and trained together with the language model.

Further, as shown in FIG4 , training a multi-design task adapter based on the Transformer model includes:

1. Multi-design task adapter: The multi-design task adapter consists of N Transformer layers, which are used to capture complex patterns and dependencies in the input data. In addition, two projection layers are added to convert the high-dimensional output of the Transformer layer into a low-dimensional representation adapted to a specific task. The multi-design task adapter is inserted as a lightweight additional component, which is located after each Transformer layer and is specifically used to introduce domain-specific knowledge without modifying the weights of the main model, thereby maintaining the generalization ability of the pre-trained model.

2. Knowledge injection mechanism: The knowledge related to each design task (design task dataset and product design domain knowledge) is encoded separately and injected into a specific multi-design task adapter through pre-training, rather than directly integrated into the main model. This allows for the flexible addition and update of customized knowledge for different tasks (such as product demand analysis, user behavior analysis) without interfering with the model's infrastructure.

3. Parameter freezing and continuous learning: The original parameters of the pre-trained model are frozen to ensure the stability and consistency of the model. The multi-design task adapter is pre-trained independently and focuses on learning the discriminative features of specific fields, which enables the model to continuously absorb without interfering with the basic model.

4. Fusion module: The structure of the fusion layer is an attention module. The query is the output of the fully connected layer in the original Transformer layer, and the value and key are the outputs of the multi-design task adapters corresponding to each task. First, a linear mapping is performed, then the query and key are dot-multiplied and normalized, and then the output is used to weight the outputs of the multi-design task adapters from different tasks.

Furthermore, as shown in FIG5 , a large language model based on multi-task adaptation and knowledge enhancement is constructed, including:

1. Demand analysis and knowledge integration

Define the scope of the field: including product design principles, market trend analysis, user behavior research, materials, production technology, and sales strategy.

Knowledge resource collection: relevant books, papers, industry reports, design specifications, case study materials (in word, ppt, txt, etc. format).

Knowledge structuring: The collected knowledge is sorted and classified, and part of it forms a structured knowledge graph to facilitate subsequent indexing and retrieval.

2. Construction of semantic retrieval module under LangChain framework

Data preprocessing: Clean, segment, and remove stop words from the text in the knowledge base to prepare for constructing vector representation.

Vectorized representation: Use the M3E (Moka Massive Mixed Embedding) and TransE models to convert text into points in a high-dimensional vector space to calculate similarity.

Build an index: Use tools such as FAISS to build an efficient approximate nearest neighbor index based on vector representation to speed up the retrieval process.

Retrieval logic implementation: Through the retrieval algorithm, combined with keyword matching and vector similarity search, it is ensured that both directly related knowledge and semantically similar content can be captured.

3. LangChain integration and information transmission

Integration with LangChain: LangChain is a framework that facilitates the integration of language models with various data sources, tools, and services. Through the LangChain interface, the retrieved information is organized into a format suitable for large language models to understand.

Context Management: Using LangChain’s ContextualRetriever, the most relevant information fragments are dynamically extracted from the knowledge base based on user queries as the input context of the large language model.

As an implementation method of an embodiment of the present invention, in step S3, the product design task toolset includes: competitive product analysis, design drawing generation, professional knowledge retrieval and design solution generation; the product design task toolset is called based on a large language model with multi-task adaptation and knowledge enhancement to obtain a product design generation agent, and the product design generation agent includes: text description of the design solution, generation of design drawings, and sales statistical analysis.

As an implementation method of an embodiment of the present invention, in step S4, the product design generation agent responds to the design query or special request immediately raised by the user, and adjusts the product design plan in real time according to the user's feedback; this interactive design helps to narrow the gap between user expectations and the final product, and enhances user participation and design satisfaction.

Embodiment 2:

The embodiment of the present invention further provides a product intelligent design system based on domain knowledge enhanced large language model, including:

An acquisition device, used for acquiring a design task data set in the field of product design;

A training device, used to obtain a large language model based on multi-task adaptation and knowledge enhancement according to a design task dataset and product design domain knowledge;

A generating device for obtaining a product design generating agent through a product design task tool set according to a large language model based on multi-task adaptation and knowledge enhancement;

The design device is used to adjust the product design scheme in real time according to user feedback through the product design generation agent.

As an implementation mode of an embodiment of the present invention, the design task data set includes: design drawings, product specifications, market analysis reports, consumer feedback and comments, material databases, and manufacturing and process parameters; the product design domain knowledge includes: knowledge graphs and expert knowledge; the product design task toolset includes: competitive product analysis, design drawing generation, professional knowledge retrieval, and design solution generation.

As an implementation of an embodiment of the present invention, the processing device includes:

A training unit, used to train a multi-design task adapter based on the Transformer model according to the design task dataset and product design domain knowledge;

A construction unit is used to build a large language model based on multi-task adaptation and knowledge enhancement based on semantic retrieval according to the trained Transformer-based multi-design task adapter and product design domain knowledge.

As an implementation manner of the embodiment of the present invention, the product is a range hood.

The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims (8)

1. A product intelligent design method based on domain knowledge enhanced large language model, characterized by comprising: Step S1, obtaining a design task dataset in the product design field; Step S2: obtaining a large language model based on multi-task adaptation and knowledge enhancement according to the design task dataset and product design domain knowledge; Step S3, obtaining a product design generation agent through a product design task tool set according to a large language model based on multi-task adaptation and knowledge enhancement; Step S4: The product design generation agent adjusts the product design plan in real time according to user feedback. 2. The product intelligent design method based on domain knowledge enhanced large language model as described in claim 1 is characterized in that the design task data set includes: design drawings, product specifications, market analysis reports, consumer feedback and comments, material databases, and manufacturing and process parameters; product design domain knowledge includes: knowledge graphs and expert knowledge; the product design task toolset includes: competitive product analysis, design drawing generation, professional knowledge retrieval, and design solution generation. 3. The product intelligent design method based on domain knowledge enhanced large language model as claimed in claim 2, characterized in that step S2 comprises: According to the design task dataset and product design domain knowledge, train a multi-design task adapter based on the Transformer model; According to the trained Transformer-based multi-design task adapter and product design domain knowledge, a large language model based on multi-task adaptation and knowledge enhancement is constructed based on semantic retrieval. 4. The product design method based on multi-task adaptation and knowledge enhancement as described in claim 3, characterized in that the product is a range hood. 5. A product intelligent design system based on domain knowledge enhanced large language model, characterized by comprising: An acquisition device, used for acquiring a design task data set in the field of product design; A processing device, for obtaining a large language model based on multi-task adaptation and knowledge enhancement according to a design task data set and product design domain knowledge; A generating device for obtaining a product design generating agent through a product design task tool set according to a large language model based on multi-task adaptation and knowledge enhancement; The design device is used to adjust the product design scheme in real time according to user feedback through the product design generation agent. 6. The product intelligent design system based on domain knowledge enhanced large language model as described in claim 5 is characterized in that the design task data set includes: design drawings, product specifications, market analysis reports, consumer feedback and comments, material database and manufacturing and process parameters; product design domain knowledge includes: knowledge graph and expert knowledge; product design task tool set includes: competitive product analysis, design drawing generation, professional knowledge retrieval and design solution generation. 7. The product intelligent design system based on domain knowledge enhanced large language model as claimed in claim 6, characterized in that the processing device comprises: A training unit, used to train a multi-design task adapter based on the Transformer model according to the design task dataset and product design domain knowledge; A construction unit is used to build a large language model based on multi-task adaptation and knowledge enhancement based on semantic retrieval according to the trained Transformer-based multi-design task adapter and product design domain knowledge. 8. The product intelligent design system based on domain knowledge enhanced large language model as described in claim 7, characterized in that the product is a range hood.