Intelligent equipment fault question-answering method and system based on knowledge graph enhancement

Abstract

The invention discloses an equipment fault intelligent question-answering method and system based on knowledge graph enhancement. The method comprises the following steps: constructing a knowledge graph of a professional field; developing a natural language processing module; designing an enhanced generation model; designing an interactive question and answer interface; the system comprises a user interaction layer, a language understanding processing layer, a knowledge graph query reasoning layer, a question and answer generation layer and a data storage management layer. Through the natural language processing technology, the system can quickly understand the query intention of a user, and the manual information input and interpretation time in traditional fault diagnosis is shortened, so that the overall fault response speed is increased; the large language model has deep learning and generalization capabilities, various complex and unknown fault conditions can be processed, accurate diagnosis and operation suggestions are provided, and the potential capability of the system for processing new fault types is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of intelligent question-answering systems, and in particular to a method and system for intelligent question-answering of equipment failures based on knowledge graph enhancement. Background Art

[0002] At present, equipment fault diagnosis mainly relies on expert experience and traditional fault code query methods, which have problems such as long diagnosis cycle and low accuracy. With the development of artificial intelligence technology, intelligent question-answering systems based on knowledge graphs have gradually become an effective means to solve these problems. With the improvement of industrial automation, the complexity of equipment continues to increase, and higher requirements are placed on the rapid diagnosis and maintenance of equipment faults. Traditional fault diagnosis methods rely on the experience of professionals, which is inefficient and prone to errors. In recent years, the development of intelligent question-answering systems has provided new solutions for automated fault diagnosis. However, existing intelligent question-answering systems still have many problems in understanding complex queries and providing accurate answers, such as: limitations in query understanding, natural language processing technology may not be sufficient to accurately parse complex or non-standard queries, especially when faced with professional terms or abbreviations in specific industries; answer accuracy issues, if the system's knowledge base is not updated in a timely manner or is not comprehensive enough, it may cause the intelligent question-answering system to provide outdated or inaccurate information; insufficient generalization ability, the intelligent question-answering system may perform well on specific fault cases, but when faced with novel or rare problems, it lacks generalization ability and has difficulty handling unseen situations, poor interactive experience, users may need to try different expressions multiple times to get useful answers, which reduces efficiency and increases user frustration, knowledge updating and maintenance difficulties: as devices are continuously iterated and updated, the knowledge base of the intelligent question-answering system needs to be continuously updated, and rapidly changing information makes maintenance a challenge; insufficient processing of deep reasoning, existing intelligent question-answering systems may have difficulty handling complex problems that require deep analysis and multi-level reasoning. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a method and system for intelligent question answering of equipment failure based on knowledge graph enhancement.

[0004] Technical solution: The device fault intelligent question-answering method based on knowledge graph enhancement described in the present invention comprises the following steps:

[0005] (1) Constructing knowledge graphs in professional fields;

[0006] (2) Develop a natural language processing module;

[0007] (3) Design enhanced generative models;

[0008] (4) Design and implement an interactive question-and-answer interface.

[0009] Furthermore, the step (1) comprises:

[0010] (1.1) Collect data from user manuals, maintenance records and expert experience through data acquisition;

[0011] (1.2) Use natural language processing technology to perform entity recognition and relationship extraction to build a preliminary knowledge structure;

[0012] (1.3) Knowledge fusion technology is used to integrate information from different sources, convert the data into a structured graph form, group related entities and relations into text blocks, and then use a word-to-vector encoder to generate vector representations.

[0013] Furthermore, the natural language processing module in step (2) includes multiple submodules, which use prompt engineering to guide the large language model to extract semantic entities from the text, and then extract keywords from the text to construct a prompt engineering template: "instruction + context + question + output", that is:

[0014] Prompt=Instruction+Context+Question+Output Format (1)

[0015] Instruction is the user's instruction, Context provides user-specific context information, Question is the user's question, and Output Format is the output format.

[0016] Furthermore, the step (2) includes embedding a large information block of the entity, fault, and solution of the equipment into a vector representation to realize vector search, using the tree structure of the knowledge graph to traverse, and for each failure cause, failure impact, and failure measure node of the equipment entity, performing a depth-first search to obtain connected related nodes, and then using the text-to-vector encoding model to calculate the vector embedding.

[0017] Furthermore, the step (3) of designing an enhanced generation model to retrieve information using a graphical query language and vector search of a knowledge graph includes the following steps:

[0018] (3.1) Obtain user questions, use prompt learning technology to guide the large language model to understand the questions, and obtain the question-answering intent.

[0019] (3.2) Based on the keywords contained in the question-answering intent, a Cypher query statement is generated to match the fault operation and maintenance knowledge subgraph required to answer the question from the fault operation and maintenance knowledge graph.

[0020] (3.3) Use prompt learning technology to guide the large language model to learn the fault operation and maintenance knowledge subgraph, and then generate answers to the questions and return them to the user.

[0021] Furthermore, the step (3) of enhancing the generation model includes:

[0022] A subgraph query is generated from the large language model based on entity keyword retrieval. If the result of the subgraph query is obtained, the information retrieval process is completed and the information is ready to be processed and presented to the user. If the subgraph query does not produce any results, the proposed framework will perform a semantic similarity query on the graph node vector storage, compare the vector representation of the user query with the node vector in the knowledge base, and complete the semantic similarity between the user input and the documents in the corpus by calculating the cosine similarity between the two vectors.

[0023] Reorder the context and select the most relevant documents to recommend to the large language model for integrated output, thereby achieving retrieval enhancement and generation results. The calculation formula for cosine similarity is:

[0024]

[0025] Among them, ||c i || represents the feature vector c of the user's query sentence i The vector norm of ||j k ||Feature vector j representing the text vector of the equipment operation and maintenance knowledge base k The vector norm of the cosine similarity is between -1 and 1, where 1 means that the two vectors are in exactly the same direction, 0 means that they are perpendicular to each other and have no similarity, and -1 means that they are in completely opposite directions. Here, the absolute value is taken so that the cosine similarity is between 0 and 1;

[0026] The retrieval results of subgraph query and vector semantic query are combined for weighted fusion. The mixed similarity is obtained by weighting the similarities of the two methods. The weighting parameter α is used to determine the weight of each algorithm and integrate the re-output of the similarity results.

[0027] Hybrid_Similarity=(1-a)*keyword_similarity+a*vector_similarity(3)

[0028] The value of α is between 0 and 1, α=1: pure vector search, α=0: pure keyword search, keyword_similarity is the text similarity of subgraph search similar to the keyword, and vector_similarity is the text similarity of search embedding similar to query embedding.

[0029] Furthermore, the step (4) of designing and implementing the fault question and answer interactive interface includes:

[0030] Adaptive knowledge retrieval method using LLM ranking ability filters out noise information, directly guides LLM to re-rank the sampled knowledge through hint engineering, and only retrieves the first k triples, given a question Q and a coarse-grained knowledge subgraph G based on paths and neighbors. Q , create prompt word pl by filling in predefined template:

[0031] p1=Template(Q,G Q ,k) (4)

[0032] Then, using p1 as input, prompt the LLM to obtain the self-retrieved knowledge G′ Q :

[0033] G′ Q =LLM(p1) (5)

[0034] Finally, construct and fill the predefined template to create prompt words p2, G′ Q Then provide it to LLM for reasoning analysis and get the predicted answer A:

[0035] p2 = Template(Q, G′ Q ) (6)

[0036] A=LLM(p2)(7)

[0037] Among them, Q represents the problem, G′ Q Represents fine-grained knowledge through question Q and fine-grained knowledge G′ Q , construct and fill in the predefined template to create prompt word p2.

[0038] Furthermore, the step (4) performs semantic analysis based on the question input by the user, and performs entity query in the system interface based on the entity name of the device. After comparing the entities and relationships related to the query entity, it can be seen that the system has processed and reprocessed the reasoning and analysis process in the background, thereby performing question and answer responses and returning accurate and clear query results.

[0039] The equipment fault intelligent question-answering system based on knowledge graph enhancement of the present invention comprises:

[0040] User interaction layer: used to provide an interface for users to input fault descriptions, supporting multiple interaction modes;

[0041] Language understanding processing layer: used to use a large language model to understand the semantics of user input and extract key information;

[0042] Knowledge graph query reasoning layer: used to query related entities and relationships in the knowledge graph based on the understanding results, and determine possible fault causes and solutions through logical reasoning;

[0043] Question and answer generation layer: used to generate natural language question and answer responses based on query and reasoning results, and provide users with fault diagnosis and maintenance suggestions;

[0044] Data storage management layer: used to store and manage knowledge graph data, model training data, and user interaction data.

[0045] Furthermore, the reasoning process of the knowledge graph query reasoning layer includes:

[0046] Obtain user questions, use prompt learning technology to guide the large language model to understand the questions, and obtain the question-answering intent; generate Cypher query statements based on the keywords contained in the question-answering intent, and match the fault operation and maintenance knowledge subgraph required to answer the question from the fault operation and maintenance knowledge graph; use prompt learning technology to guide the large language model to learn the fault operation and maintenance knowledge subgraph, and then generate the answer to the question and return it to the user.

[0047] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: Through natural language processing technology, the system can quickly understand the user's query intention, reducing the time for manual information entry and interpretation in traditional fault diagnosis, thereby improving the overall fault response speed; the large language model has the ability of deep learning and generalization, can handle various complex and unknown fault conditions, provide accurate diagnosis and operation suggestions, and enhance the system's potential ability to handle new fault types; users can interact with the system through natural language, which lowers the operating threshold and improves the experience of non-professionals in getting help by themselves when equipment fails; the automated fault diagnosis and question-answering system reduces the dependence on professional operation and maintenance personnel, which can significantly reduce manpower cost, and also reduces the additional maintenance costs caused by human errors; the system can learn historical fault data and solutions, continuously optimize knowledge graphs and language models, promote knowledge accumulation and sharing, and make operation and maintenance knowledge management more efficient and centralized; fast and accurate fault diagnosis helps to eliminate equipment failures in a timely manner, reduce production line downtime, ensure the continuity and stability of the production process, and prevent potential safety hazards caused by failures; the data generated during the operation of the system can be further analyzed to discover potential defects and maintenance needs of the equipment, and support the implementation of predictive maintenance strategies; it provides technical support for intelligent manufacturing, promotes the improvement of industrial automation and intelligence levels, and meets the needs of transformation and upgrading of the modern manufacturing industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is the overall architecture diagram of the equipment failure intelligent question-answering method based on knowledge graph enhancement;

[0049] Figure 2 This is a module diagram of the equipment failure intelligent question-answering system based on knowledge graph enhancement;

[0050] Figure 3 Flowchart of query reasoning based on knowledge graph retrieval enhancement. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0052] like Figure 1 As shown, the device fault intelligent question-answering method based on knowledge graph enhancement of the present invention comprises the following steps:

[0053] (1) Constructing a knowledge graph in a professional field: The knowledge graph is the core of the present invention, which contains a wealth of entities and relationships, such as equipment models, component names, fault types, influencing factors and solutions. First, through data acquisition, data is collected from user manuals, maintenance records and expert experience. Next, natural language processing technology is used to perform entity recognition and relationship extraction to build a preliminary knowledge structure. Then, knowledge fusion technology is used to integrate information from different sources, eliminate ambiguity, and ensure the accuracy and consistency of the knowledge graph. The data is converted into a structured graph form to facilitate semantic query and analysis. At the same time, related entities and relationships are combined into text blocks, and then a word-to-vector encoder is used to generate vector representations. These vector embeddings are used for vector search in KG to improve the accuracy and efficiency of information retrieval.

[0054] (2) Develop a natural language processing module: This module is responsible for parsing the user's natural language query. It contains multiple submodules, such as a tokenizer, a part-of-speech tagger, and a dependency parser. These tools work together to extract meaningful information from the text. For example, when a user enters "What should I do if the engine overheats?", the system can recognize that "engine" is the subject of the study, "overheating" is the fault phenomenon, and "what to do" is the solution.

[0055] The natural language processing module includes multiple sub-modules. It uses prompt engineering to guide the large language model to extract semantic entities from the text, and then extract keywords from the text to build a prompt engineering template: "instruction + context + question + output", that is:

[0056] Prompt=Instruction+Context+Question+Output Format (1)

[0057] Here, Instruction is the user's instruction, Context provides user-specific contextual information, Question is the user's question, and Output Format is the output format. For example, when the user inputs "What should I do if the engine overheats?", after being analyzed and processed by the large model, it can identify the output "engine" as a device entity, "overheating" as a fault phenomenon, and "what to do" to find a solution.

[0058] At the same time, although there are complex reasoning methods within the knowledge graph, they cannot scale with the size of the ontology, making it difficult to retrieve deeper combined information. To solve this problem, we embed larger information blocks of node data such as the entity, fault, and solution of the device into vector representations to achieve vector search, thereby using reasoning paths to achieve retrieval queries for complex information. Here, the tree structure of the knowledge graph is mainly used to traverse. For each device entity's FailureCause, FailureEffect, and FailureMesaure nodes, a depth-first search (DFS) is performed to obtain connected related nodes, which ensures the collection of all branch nodes. Using the generated text blocks, the text-to-vector encoding model (SentenceTransformer) is used to calculate the vector embedding.

[0059] (3) Designing an enhanced generation model: The enhanced generation model is one of the innovations of the present invention. Knowledge Graph Retrieval-Augmented Generation (KG-RAG) is a technical framework that combines knowledge graph retrieval with a large language model to improve the performance of the question-answering system, enabling the model to focus on the most relevant part of the knowledge graph to generate more accurate answers.

[0060] The enhanced generation model includes generating a subgraph query based on entity keyword retrieval from a large language model. If the result of the subgraph query is obtained, the information retrieval process is completed and the information is ready to be processed and presented to the user. If the subgraph query does not produce any results, the proposed framework will perform a semantic similarity query on the graph node vector storage, compare the vector representation of the user query with the node vector in the knowledge base, and complete the semantic similarity between the user input and the document in the corpus by calculating the cosine similarity between the two vectors.

[0061] Reorder the context and select the most relevant documents to recommend to the large language model for integrated output, thereby achieving retrieval enhancement and generation results. The calculation formula for cosine similarity is:

[0062]

[0063] Among them, ||c i || represents the feature vector c of the user's query sentence i The vector norm of ||j k ||Feature vector j representing the text vector of the equipment operation and maintenance knowledge base k The vector norm of the cosine similarity is between -1 and 1, where 1 means that the two vectors are in exactly the same direction, 0 means that they are perpendicular to each other and have no similarity, and -1 means that they are in completely opposite directions. Here, the absolute value is taken so that the cosine similarity takes a value between 0N1;

[0064] The retrieval results of subgraph query and vector semantic query are combined for weighted fusion. The mixed similarity is obtained by weighting the similarities of the two methods. The weighting parameter α is used to determine the weight of each algorithm and integrate the re-output of the similarity results.

[0065] Hybrid_Similarity=(1-a)*keyword_similarity+a*vector_similarity (3)

[0066] The value of α is between 0 and 1, α=1: pure vector search, α=0: pure keyword search, keyword_similarity is the text similarity of subgraph search similar to the keyword, and vector_similarity is the text similarity of search embedding similar to query embedding.

[0067] Key technical methods for query reasoning based on knowledge graph retrieval enhancement, such as Figure 3 As shown:

[0068] (3.1) Obtain user questions, use prompt learning technology to guide the large language model to understand the questions, and obtain the question-answering intent.

[0069] (3.2) Based on the keywords contained in the question-answering intent, a Cypher query statement is generated to match the fault operation and maintenance knowledge subgraph required to answer the question from the fault operation and maintenance knowledge graph.

[0070] (3.3) Use prompt learning technology to guide the large language model to learn the fault operation and maintenance knowledge subgraph, and then generate answers to the questions and return them to the user.

[0071] We propose an adaptive knowledge retrieval method that leverages the ranking capability of LLM to filter out noisy information. Through hint engineering, we directly guide LLM to re-rank the sampled knowledge and retrieve only the top k triples. Given a question Q and a coarse-grained knowledge subgraph G based on paths and neighbors, Q , create prompt word p1 by filling in the predefined template:

[0072] p1=Template(Q,G Q ,k) (4)

[0073] Then, using p1 as input, prompt the LLM to obtain the self-retrieved knowledge G′ Q :

[0074] G′ Q =LLM(p1) (5)

[0075] Finally, construct and fill the predefined template to create prompt words p2, G′ Q Then provide it to LLM for reasoning analysis and get the predicted answer A:

[0076] p2 = Template(Q, G′ Q ) (6)

[0077] A=LLM(p2) (7)

[0078] Among them, Q represents the problem, G′ Q Represents fine-grained knowledge through question Q and fine-grained knowledge G′ Q , construct and fill in the predefined template to create prompt word p2.

[0079] Based on the questions input by the user, semantic analysis is performed, and entity queries are performed in the system interface based on the entity name of the device. After comparing the entities and relationships related to the query entities, it can be seen that the system has processed the reasoning and analysis process in the background, so as to respond to questions and answers and return accurate and clear query results.

[0080] (4) Implement an interactive question-and-answer interface: The interface design should be simple and intuitive, and support multiple interaction modes, such as text input, voice recognition, and other multimodal terminal applications, to meet the needs of different users.

[0081] The device module of the equipment fault intelligent question-answering system based on knowledge graph enhancement described in the present invention is as follows: Figure 2 As shown:

[0082] User interaction layer: provides an interface for users to input fault descriptions and supports multiple interaction methods, such as text input and voice recognition.

[0083] Language understanding processing layer: Use a large language model to understand the semantics of user input and extract key information, such as device type, fault phenomenon, fault conditions, etc.

[0084] Knowledge graph query reasoning layer: Based on the understanding results, query the relevant entities and relationships in the knowledge graph, and determine the possible causes and solutions of the failure through logical reasoning.

[0085] Question and answer generation layer: Based on the query and reasoning results, it generates question and answer responses in natural language to provide users with fault diagnosis and maintenance suggestions.

[0086] Data storage management layer: stores and manages knowledge graph data, model training data, user interaction data, etc. to ensure data consistency and security.

[0087] The combination of this method and system framework will be able to realize an efficient and intelligent equipment failure operation and maintenance knowledge question and answer system, which will help improve equipment maintenance efficiency, reduce downtime, reduce maintenance costs, and enhance the reliability and stability of intelligent manufacturing systems.

[0088] The innovation of this invention lies in its unique method of combining a large language model with a graph database, which fully utilizes the advantages of a large language model in semantic understanding and knowledge reasoning, as well as the powerful capabilities of a graph database in data organization and query efficiency. In addition, through the application of prompt learning technology, the performance and flexibility of the system in dealing with complex problems are further improved. This comprehensive technical solution provides a new intelligent and automated solution for the field of equipment fault operation and maintenance, significantly improves the speed and quality of fault handling, reduces operation and maintenance costs, and has extremely high practical value and market potential.

[0089] The knowledge graph retrieval enhancement technology proposed in this invention is particularly suitable for vertical fields that require high accuracy, such as medicine, legal consulting, and internal knowledge management of enterprises. In this case, the accuracy of the model is crucial, and the knowledge graph can provide the necessary professional knowledge support. The knowledge graph retrieval enhancement technology is a technology that combines the powerful generation capability of a large language model with the precise information retrieval capability of a knowledge graph. It can significantly improve the accuracy and reliability of the question-answering system while maintaining the model generation capability.

Claims

1. An intelligent question-answering method for equipment failure based on knowledge graph enhancement, characterized in that: The steps include: (1) Constructing knowledge graphs in professional fields; (2) Develop a natural language processing module; (3) Design enhanced generative models; (4) Design and implement an interactive question-and-answer interface.

2. The device fault intelligent question-answering method based on knowledge graph enhancement according to claim 1 is characterized in that: The step (1) comprises: (1.1) Collect data from user manuals, maintenance records and expert experience through data acquisition; (1.2) Use natural language processing technology to perform entity recognition and relationship extraction to build a preliminary knowledge structure; (1.3) Knowledge fusion technology is used to integrate information from different sources, convert the data into a structured graph form, group related entities and relations into text blocks, and then use a word-to-vector encoder to generate vector representations.

3. The device fault intelligent question-answering method based on knowledge graph enhancement according to claim 1 is characterized in that: The natural language processing module in step (2) includes multiple submodules, which use prompt engineering to guide the large language model to extract semantic entities from the text, and then extract keywords from the text to construct a prompt engineering template: "instruction + context + question + output", that is: Prompt=Instruction+Context+Question+Output Format (1) Among them, Instruction is the user's instruction, Context provides user-specific context information, Question is the user's question, and Output Format is the output format.

4. The device fault intelligent question-answering method based on knowledge graph enhancement according to claim 1 is characterized in that: The step (2) includes embedding a large information block of the entity, fault, and solution of the equipment into a vector representation to realize vector search, using the tree structure of the knowledge graph to traverse, and performing a depth-first search for each failure cause, failure impact, and failure measure node of each equipment entity to obtain connected related nodes, and then using the text-to-vector encoding model to calculate the vector embedding.

5. The device fault intelligent question-answering method based on knowledge graph enhancement according to claim 1 is characterized in that: The step (3) of designing an enhanced generation model to use the graphical query language of the knowledge graph and vector search to retrieve information includes the following steps: (3.1) Obtain user questions, use prompt learning technology to guide the large language model to understand the questions, and obtain the question-answering intent. (3.2) Based on the keywords contained in the question-answering intent, a Cypher query statement is generated to match the fault operation and maintenance knowledge subgraph required to answer the question from the fault operation and maintenance knowledge graph. (3.3) Use prompt learning technology to guide the large language model to learn the fault operation and maintenance knowledge subgraph, and then generate answers to the questions and return them to the user.

6. The device fault intelligent question-answering method based on knowledge graph enhancement according to claim 1 is characterized in that: The step (3) of enhancing the generation model comprises: A subgraph query is generated from the large language model based on entity keyword retrieval. If the result of the subgraph query is obtained, the information retrieval process is completed and the information is ready to be processed and presented to the user. If the subgraph query does not produce any results, the proposed framework will perform a semantic similarity query on the graph node vector storage, compare the vector representation of the user query with the node vector in the knowledge base, and complete the semantic similarity between the user input and the documents in the corpus by calculating the cosine similarity between the two vectors. Reorder the context and select the most relevant documents to recommend to the large language model for integrated output, thereby achieving retrieval enhancement and generation results. The calculation formula for cosine similarity is: Among them, ||c i || represents the feature vector c of the user's query sentence i The vector norm of ||j k ||Feature vector j representing the text vector of the equipment operation and maintenance knowledge base k The vector norm of the cosine similarity is between -1 and 1, where 1 means that the two vectors are in exactly the same direction, 0 means that they are perpendicular to each other and have no similarity, and -1 means that they are in completely opposite directions. Here, the absolute value is taken so that the cosine similarity is between 0 and 1; The retrieval results of subgraph query and vector semantic query are combined for weighted fusion. The mixed similarity is obtained by weighting the similarities of the two methods. The weighting parameter α is used to determine the weight of each algorithm and integrate the re-output of the similarity results. Hybrid_Similarity=(1-a)*keyword_similarity+a*vector_similarity(3) The value of α is between 0 and 1, α=1: pure vector search, α=0: pure keyword search, keyword_similarity is the text similarity of subgraph search similar to the keyword, and vector_similarity is the text similarity of search embedding similar to query embedding.

7. The device fault intelligent question-answering method based on knowledge graph enhancement according to claim 1 is characterized in that: The step (4) of designing and implementing the fault question and answer interactive interface includes: Adaptive knowledge retrieval method using LLM ranking ability filters out noise information, directly guides LLM to re-rank the sampled knowledge through hint engineering, and only retrieves the first k triples, given a question Q and a coarse-grained knowledge subgraph G based on paths and neighbors. Q , create prompt word p1 by filling in the predefined template: pl=Template(Q,G Q ,k) (4) Then, using p1 as input, prompt the LLM to obtain the self-retrieved knowledge G′ Q : G′ Q =LLM(p1) (5) Finally, construct and fill the predefined template to create prompt words p2, G′ Q Then provide it to LLM for reasoning analysis and get the predicted answer A: p2=Template(Q,G′ Q ) (6) A=LLM(p2) (7) Among them, Q represents the problem, G′ Q Represents fine-grained knowledge through question Q and fine-grained knowledge G′ Q , construct and fill in the predefined template to create prompt word p2.

8. The device fault intelligent question-answering method based on knowledge graph enhancement according to claim 1 is characterized in that: The step (4) performs semantic analysis based on the question input by the user, and performs entity query in the system interface based on the entity name of the device. After comparing the entities and relationships related to the query entity, it can be seen that the system has processed and reprocessed the reasoning and analysis process in the background, thereby performing question and answer responses and returning accurate and clear query results.

9. An intelligent question-answering system for equipment failure based on knowledge graph enhancement, characterized in that: include: User interaction layer: used to provide an interface for users to input fault descriptions, supporting multiple interaction modes; Language understanding processing layer; Used to use large language models to understand the semantics of user input and extract key information; Knowledge graph query reasoning layer; Used to query relevant entities and relationships in the knowledge graph based on the understanding results, and determine possible fault causes and solutions through logical reasoning; Question and answer generation layer: used to generate natural language question and answer responses based on query and reasoning results, and provide users with fault diagnosis and maintenance suggestions; Data storage management layer: used to store and manage knowledge graph data, model training data, and user interaction data.

10. The equipment fault intelligent question-answering system based on knowledge graph enhancement according to claim 9 is characterized in that: The reasoning process of the knowledge graph query reasoning layer includes: Obtain user questions, use prompt learning technology to guide the large language model to understand the questions, and obtain the question-answering intent; generate Cypher query statements based on the keywords contained in the question-answering intent, and match the fault operation and maintenance knowledge subgraph required to answer the question from the fault operation and maintenance knowledge graph; use prompt learning technology to guide the large language model to learn the fault operation and maintenance knowledge subgraph, and then generate the answer to the question and return it to the user.

Images