Search

CN-121996759-A - Knowledge graph enhancement-based manufacturing process knowledge multi-path search question-answering method

CN121996759ACN 121996759 ACN121996759 ACN 121996759ACN-121996759-A

Abstract

The invention discloses a knowledge graph enhancement-based manufacturing process knowledge multi-path search question-answering method, which comprises the steps of firstly carrying out knowledge structured extraction on a manufacturing process document, then constructing a multi-mode index and a knowledge graph, then intelligently analyzing query intention, then carrying out knowledge graph enhancement multi-path mixed search, then carrying out deep reordering and context enhancement, and finally completing answer generation based on enhancement context. The invention overcomes the defect of insufficient understanding of the traditional analysis mode on the complex document structure, lays a solid structural foundation for the complex technical reasoning of the cross-document, and remarkably improves the reasoning capacity and the answer accuracy of the system.

Inventors

  • Geng Ruibin
  • LIU ZHIJUN
  • ZHOU RAN
  • HUANG YONGSHENG
  • WANG BOWEI
  • LI JINGYUAN

Assignees

  • 西北工业大学
  • 中国航发动力股份有限公司

Dates

Publication Date
20260508
Application Date
20260108

Claims (9)

  1. 1. The knowledge graph enhancement-based manufacturing process knowledge multi-path search question-answering method is characterized by comprising the following steps of: step 1, knowledge structured extraction of a manufacturing process document; Converting the unstructured process document into a structured knowledge entity; step 2, constructing a multi-mode index and a knowledge graph; Establishing a composite index structure supporting multiple retrieval modes, and constructing logic association of knowledge of a knowledge graph expression process; Step 3, intelligent analysis of query intention; Deconstructing a user's natural language query into a structured search instruction; step 4, multi-path mixed retrieval with enhanced knowledge graph; recalling the related process knowledge through a plurality of parallel retrieval paths; step 5, depth reordering and context enhancement; Sequencing the candidate knowledge and constructing a structured context; Step 6, generating an answer based on the enhanced context; an answer is generated based on the enhanced context using the language generation model.
  2. 2. The knowledge graph enhancement-based manufacturing process knowledge multi-path search question-answering method according to claim 1, wherein the step 1 is specifically: Analyzing and partitioning the technical document, and dividing the technical document into independent text units according to a hierarchical structure of the document; step 1.2, carrying out knowledge type identification on the text unit, and determining the process knowledge category contained in the text unit; Step 1.3, extracting structured knowledge elements by adopting a corresponding extraction strategy according to the identified knowledge type; And 1.4, verifying and standardizing the extracted structured data to form standardized technological knowledge items.
  3. 3. The method of claim 2, wherein the process knowledge categories include at least one or more of a term definition category, a process parameter category, a process step category, a constraint category, and a causal relationship category.
  4. 4. The knowledge graph enhancement-based manufacturing process knowledge multi-path search question-answering method according to claim 2, wherein the step 2 is specifically: Step 2.1, creating a document context node in a graph database, and recording meta-information of the document; Step 2.2, instantiating the structured data extracted in the step 1 into process knowledge nodes, and establishing association relations with corresponding document context nodes; step 2.3, carrying out entity normalization processing on core elements in the process knowledge nodes to create concept nodes; step 2.4, establishing a semantic relation edge between a process knowledge node and a concept node according to the internal logic of the process knowledge to form a process knowledge map; Step 2.5, converting text description of the process knowledge nodes into vector representation by utilizing a semantic embedding technology, and constructing a semantic vector index; and 2.6, constructing an accurate matching index aiming at key fields in the process knowledge nodes.
  5. 5. The knowledge-graph-enhancement-based manufacturing process knowledge multi-path search question and answer method according to claim 4, wherein the semantic relationship side at least comprises one or more of a definition relationship, a parameter relationship, a step relationship, a causal relationship and a constraint relationship.
  6. 6. The knowledge graph enhancement-based manufacturing process knowledge multi-path search question-answering method according to claim 4, wherein the step 3 is specifically: step 3.1, receiving natural language query input by a user; Performing intent analysis on the query, and identifying the query type, the query entity and the query constraint; extracting generalized search terms for fuzzy matching and entity names for accurate matching from the query; And 3.4, packaging the extracted search terms, the entity and the query type into a structured search instruction.
  7. 7. The knowledge graph-enhanced manufacturing process knowledge multi-path search question-answering method according to claim 6, wherein the step 4 is specifically: step 4.1, searching semantic similarity based on the generalized search word, and recalling the process knowledge nodes related to the semantic; Step 4.2, carrying out keyword matching retrieval based on the accurate entity, and recalling the process knowledge node containing the specific term; step 4.3, aiming at complex inquiry comprising a plurality of entities, performing relation reasoning in a knowledge graph, searching logic paths connected with different entities, and recalling intermediate process knowledge nodes on the paths; 4.4, aiming at the condition that single entity inquiry or relation reasoning is insufficient in recall, carrying out map expansion by taking a precisely recalled node as a starting point, and carrying out neighbor node screening by combining semantic similarity; And 4.5, merging and deduplicating the process knowledge nodes recalled by the multipath retrieval to generate a candidate knowledge set.
  8. 8. The knowledge graph enhancement-based manufacturing process knowledge multi-path search question-answering method according to claim 7, wherein the step 5 is specifically: Step 5.1, obtaining complete text description of each process knowledge node in the candidate knowledge set; step 5.2, constructing text pairs of query and process knowledge; Step 5.3, calculating a correlation score between the query and the process knowledge by using the deep semantic matching model; step 5.4, sorting according to the relevance scores, and screening out highly relevant process knowledge; Step 5.5, extracting the screened process knowledge nodes and the relation edges thereof from the knowledge graph to construct an inference subgraph; And 5.6, converting the inference sub-graph into a natural language text containing knowledge content and relationship description to form an enhanced context.
  9. 9. The knowledge graph enhancement-based manufacturing process knowledge multi-path search question-answering method according to claim 8, wherein the step 6 is specifically: step 6.1, constructing a generating prompt template comprising role definition, task instructions and constraint conditions; Step 6.2, filling the enhanced context generated in the step 5 and the original query of the user into a prompt template; Step 6.3, inputting the complete prompt information into a language generation model for reasoning; And 6.4, the language generation model carries out logic fusion and reasoning on the process knowledge in the enhanced context to generate an answer containing a conclusion, a reasoning process and a knowledge source.

Description

Knowledge graph enhancement-based manufacturing process knowledge multi-path search question-answering method Technical Field The invention belongs to the technical field of knowledge graphs, and particularly relates to a knowledge graph-enhanced manufacturing process knowledge multi-path search question-answering method. Background With the deep advancement of industrial digitization and intelligent manufacturing, a large number of manufacturing process documents, including unstructured technical documents such as technical standard specifications, process flow files, operation rules, equipment maintenance manuals and the like, are accumulated in enterprises. These documents are a core source of knowledge for process design, production decisions, and troubleshooting. However, because of the high specificity, the strict logic and the complex parameter association characteristics of the manufacturing process knowledge, the conventional information retrieval method is difficult to meet the actual requirements. When facing the complex engineering problem of cross-document, technicians often need to comb multi-step causal chains, trace back the process parameter dependency relationship and conduct multi-condition constraint comparison analysis, but because of knowledge fragmentation storage, information acquisition efficiency is low, retrieval blind areas or logic faults are easy to generate, and potential risks are brought to production safety and technical decision. The existing intelligent question-answering technology mainly relies on single vector similarity retrieval or keyword matching, and although text fragments can be located, deep topological structures and logic correlations between knowledge points are difficult to capture. Keyword retrieval does not understand semantic variants, while simple vector retrieval often returns only fragmented paragraphs, lacking efficient structural expression and storage mechanisms for "entity-attribute-relationship" triples (e.g. "process parameters-numerical range-constraints") that are ubiquitous in manufacturing process knowledge. When faced with complex queries requiring multi-step logical reasoning (e.g. "what defects a certain process will produce for a certain material at a certain temperature"), the existing methods cannot effectively integrate the associated knowledge scattered in different document chapters, and the relevance ranking of the search results is often not accurate enough, resulting in the lack of strict and complete context support of the generated large model when answering the questions, which is very easy to generate "machine illusions" or generate wrong answers lacking basis. Disclosure of Invention In order to overcome the defects of the prior art, the invention provides a knowledge graph enhancement-based multi-path searching and answering method for manufacturing process knowledge, which comprises the steps of firstly carrying out knowledge structured extraction on a manufacturing process document, then constructing a multi-mode index and a knowledge graph, then intelligently analyzing a query intention, then carrying out multi-path mixed searching of the knowledge graph enhancement, then carrying out deep reordering and context enhancement, and finally completing answer generation based on the enhancement context. The invention overcomes the defect of insufficient understanding of the traditional analysis mode on the complex document structure, lays a solid structural foundation for the complex technical reasoning of the cross-document, and remarkably improves the reasoning capacity and the answer accuracy of the system. The technical scheme adopted for solving the technical problems is as follows: step 1, knowledge structured extraction of a manufacturing process document; Converting the unstructured process document into a structured knowledge entity; step 2, constructing a multi-mode index and a knowledge graph; Establishing a composite index structure supporting multiple retrieval modes, and constructing logic association of knowledge of a knowledge graph expression process; Step 3, intelligent analysis of query intention; Deconstructing a user's natural language query into a structured search instruction; step 4, multi-path mixed retrieval with enhanced knowledge graph; recalling the related process knowledge through a plurality of parallel retrieval paths; step 5, depth reordering and context enhancement; Sequencing the candidate knowledge and constructing a structured context; Step 6, generating an answer based on the enhanced context; an answer is generated based on the enhanced context using the language generation model. Preferably, the step 1 specifically includes: Analyzing and partitioning the technical document, and dividing the technical document into independent text units according to a hierarchical structure of the document; step 1.2, carrying out knowledge type identification on the text unit, and determining the proc