CN-121995876-A - GraphRAG-based size model collaborative process route generation method

CN121995876ACN 121995876 ACN121995876 ACN 121995876ACN-121995876-A

Abstract

The invention belongs to the technical field of intelligent manufacturing, and particularly relates to a method for generating a collaborative process route based on graphRAG size models, which comprises the steps of S1, receiving multimodal input data, generating multimodal prompt words oriented to forging process planning, S2, inputting the multimodal prompt words into a prompt word optimization module, encoding the multimodal prompt words into query vectors, executing mixed retrieval in a knowledge graph, then outputting enhanced prompt words by combining a chain logic reasoning mechanism, S3, inputting the enhanced prompt words into a large language model, performing global semantic reasoning and process path planning, generating structural process route data, S4, performing multimodal structural analysis on the structural process route data, and synchronously generating interpretable process documents, visual flow representations and sustainable structural process data. The method can synchronously improve the accuracy, the interpretability, the structuring degree and the integration capability with a production system of the process scheme.

Inventors

DUAN HONGJUN
YAN PING
WANG BOCHENG
LIN JUNYAO
ZHAI HONGJIN

Assignees

重庆大学

Dates

Publication Date: 20260508
Application Date: 20251230

Claims (10)

1. The method for generating the collaborative process route based on graphRAG size models is characterized by comprising the following steps of: S1, receiving multi-mode input data, carrying out modal classification on the multi-mode input data through a multi-mode identification model to obtain multi-class single-mode data, calling a corresponding prompting word template in a preset single-mode prompting word template library for each class of single-mode data, and carrying out parameter filling and rule mapping according to the content of the single-mode data to generate a plurality of single-mode prompting word fragments; S2, inputting the multi-mode prompt words generated in the step S1 into a prompt word optimization module; the prompt word optimization module is based on graphRAG architecture and depends on a pre-constructed knowledge graph of the forging field, wherein the knowledge graph comprises a plurality of process related entity nodes and semantic relation edges for representing logic constraint among entities; the prompt word optimization module encodes the multi-mode prompt word into a query vector, and performs mixed search in the knowledge graph, wherein on one hand, entity nodes with similar semantics are matched based on vector similarity, and on the other hand, path traversal and structural reasoning are performed along the semantic relation edge in the graph, so that a knowledge sub-graph related to the current process context is recalled; S3, inputting the enhanced prompt word output by the S2 into a large language model, analyzing the semantics of the enhanced prompt word by the large language model, initiating a call request to a corresponding special small model deployed in a model coordination server through a standardized function call interface according to the identified process planning task type, wherein the special small model is used for executing quality prediction, process parameter association analysis or multi-objective parameter optimization tasks and returning corresponding calculation results; s4, carrying out multi-mode structural analysis on the structural process route data generated in the step S3, extracting process element information, and synchronously generating an interpretable process document, a visual flow representation and structural process data which can be stored in a durable mode so as to integrate the process data with a production execution system.
2. The method for generating a collaborative process route based on a graphRAG size model as set forth in claim 1, wherein S1 the multimodal input data includes at least two of a forging end map, a quality requirement text, order data, equipment parameters, and material properties.
3. The method for generating a collaborative process route based on a graphRAG size model as set forth in claim 1, wherein S1, the multimodal recognition model employs a multimodal transducer architecture for classifying input data into an image modality, a natural language text modality, or a structured numerical modality.
4. The method of claim 1, wherein S1, the single-mode prompt word template library comprises image prompt word templates, text prompt word templates and numerical prompt word templates, and parameter placeholders and combination rules are preset for each template respectively for realizing parameter replacement and rule mapping.
5. The method for generating a collaborative process route based on graphRAG models of size according to claim 1, wherein in S2, a knowledge graph of a forging domain includes a plurality of types of entity nodes and semantic relationship edges connecting the entity nodes, wherein the types of the entity nodes include product types, process stages, equipment capabilities, material characteristics and quality indexes, and the semantic relationship edges are used for characterizing at least one of process constraints, material suitability or equipment compatibility.
6. The method for generating a collaborative process route based on a size model of graphRAG according to claim 5, wherein in S3, if the calculation result returned by the special small model conflicts with the process constraint condition, a retry mechanism is triggered, and the small model or the adjustment prompt word is recalled until the structural process route data meeting the constraint is generated.
7. The method for generating a collaborative process route based on a size model of graphRAG according to claim 1, wherein in S2, a chained logical reasoning mechanism is used to perform consistency check and context completion on the recalled knowledge subgraph to suppress the illusion output of the large language model.
8. The method for generating a collaborative process route based on a size model graphRAG as set forth in claim 1, wherein in S3, the model coordination server is MCP SERVER, which uniformly encapsulates and deploys all specialized small models and algorithms, provides a standardized interface conforming to the MCP specification, and supports functional tool calls or remote procedure calls.
9. The method for generating a collaborative process route based on graphRAG size models as set forth in claim 1, wherein in S3, the dedicated small models include a die forging dedicated small model and a free forging dedicated small model.
10. The method for generating a collaborative process route based on a size model of graphRAG as set forth in claim 1, wherein in S4, the interpretable process document is a printable process card, the visualization process is represented as a process flow graph rendered based on process node and flow direction relationships, and the persistently stored structured data is written to a relational database to form process database instances.

Description

GraphRAG-based size model collaborative process route generation method Technical Field The invention belongs to the technical field of intelligent manufacturing, and particularly relates to a graphRAG-based size model collaborative process route generation method. Background With the deep fusion of artificial intelligence and advanced manufacturing technology, intelligent manufacturing has become a core driving force for promoting transformation and upgrading of industrial systems. Under the background, the process planning is used as a key link in the mechanical manufacturing, and the intelligent level directly determines the product quality, the production efficiency and the resource utilization efficiency. Traditional process planning relies primarily on engineer experience or rule-based expert systems, which are difficult to address with the challenges presented by current complex product structures, multi-material systems, and high flexible production requirements. Therefore, how to construct an intelligent planning method which can integrate multi-source heterogeneous data, has the capability of deep knowledge reasoning, can generate a structured, interpretable and executable process route becomes a key technical bottleneck which needs to be broken through in the intelligent manufacturing field. In recent years, large language models (Large Language Models, LLMs) have demonstrated great potential in process knowledge modeling and aid decision making by virtue of their powerful semantic understanding and text generation capabilities. However, the process route generation by relying on a large model is still subject to significant limitations, on one hand, the large model lacks accurate modeling and computing capability of engineering elements such as process mechanism, equipment constraint, material performance and the like in the physical world, so that phantom output is easy to generate, and the generated result is unreliable in engineering practice, and on the other hand, although partial research attempts to improve the performance of the large model on specific tasks through fine tuning or prompting engineering (such as the intelligent planning method of the process flow based on the large model, which is proposed by patent CN 117436236A), the problem of knowledge deletion and logic inconsistency is not solved effectively, and the feasibility and consistency of a process scheme are difficult to ensure. Meanwhile, although a small model facing a specific process task (such as quality prediction, parameter optimization, association analysis and the like) has high-precision numerical calculation capability, the small model has weak context awareness capability and limited generalization range, and cannot independently complete a global process planning task covering multiple procedures and multiple constraints. Therefore, how to realize the efficient coordination between the semantic reasoning advantages of the large model and the accurate computing capability of the small model becomes a key point for improving the overall performance of the intelligent process generation system. At the knowledge organization level, the prior art knowledge exists in the form of unstructured documents, drawings or database records, and lacks unified semantic expression and logic association. Even if some methods attempt to introduce geometric feature matching or a rule engine (such as the automatic generation method of the technological process based on three-dimensional model comparison proposed by patent CN106529028 a), the knowledge expression is still limited to local features, and multi-dimensional explicit technological knowledge maps of covering materials, devices, procedures, constraints and the like cannot be constructed, so that cross-procedure logical reasoning, constraint propagation and context consistency verification cannot be supported. This directly results in the generated process route lacking in interpretability, robustness, and engineering executability. In addition, the output form of most of current process generation systems is still unstructured text or static table, so that the time sequence logic and the dependency relationship of the process flow are difficult to visually present, seamless integration with platforms such as an enterprise-level Manufacturing Execution System (MES), product life cycle management (PLM) and the like is also impossible, and the traceability, multiplexing and closed-loop optimization capability of process data are severely restricted. The root causes of the problems are that (1) an effective cooperative mechanism is lacking between size models, so that semantic understanding and engineering calculation are disjointed, (2) process knowledge is not structured and mapped, the reasoning depth and knowledge multiplexing efficiency of a system are limited, and (3) a standardized and structured expression form is lacking in a generated result, so that deep integration