Equipment diagnosis method and system

Abstract

The invention provides an equipment diagnosis method and system, and relates to the technical field of natural language processing, and the method comprises the steps: carrying out the extension rewriting of the problem description of an equipment fault, and obtaining the extension description of the equipment fault; decomposing the extension description to obtain a plurality of sub-problems; retrieving from the knowledge graph according to each sub-question to obtain a retrieval sub-graph corresponding to each sub-question; performing dynamic reasoning according to the retrieval sub-graph, and adjusting a retrieval path of the retrieval sub-graph in the knowledge graph; and determining a diagnosis result of the equipment fault according to the retrieval path. According to the method, the diagnosis precision is improved by utilizing the existing knowledge and relationship in the knowledge graph, and the efficiency and adaptability of the diagnosis process are also improved, so that the method can cope with various complicated and variable equipment fault conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a device diagnosis method and system. Background Art

[0002] At present, when a device fails, it is necessary to quickly and accurately locate the cause of the failure and find a reasonable solution, as well as the risk of concurrent failures of related equipment. Usually, it is necessary to rely on the experience of engineers to analyze the failure phenomenon. However, relying solely on experience requires high professionalism in equipment failure maintenance, and historical experience knowledge is easy to be lost. At the same time, the lack of scientific basis for optimization decisions can easily lead to omissions in fault analysis.

[0003] In the prior art, in order to make up for the shortcomings of manual analysis, large language models are usually used to search external knowledge bases. However, large language models mainly rely on text blocks to store data and are unable to mine invisible connections between texts. As new data increases, it may be necessary to reorganize existing text blocks or even repeatedly store multiple versions of similar content, which will increase the size of the data and affect the retrieval performance of the system. In addition, when faced with complex or semantically ambiguous queries, they are easily interfered by more noise data, thereby affecting the diagnosis of the equipment. Summary of the invention

[0004] The problem solved by the present invention is how to improve the diagnosis accuracy for equipment failure.

[0005] In order to solve the above problems, the present invention provides an equipment diagnosis method and system.

[0006] In a first aspect, the present invention provides a device diagnosis method, comprising:

[0007] Expand and rewrite the problem description of the equipment failure to obtain an expanded description of the equipment failure;

[0008] Decomposing the extended description into a plurality of sub-problems;

[0009] Searching from the knowledge graph according to each of the sub-questions to obtain a search sub-graph corresponding to each of the sub-questions;

[0010] Performing dynamic reasoning based on the retrieval subgraph to adjust the retrieval path of the retrieval subgraph in the knowledge graph;

[0011] A diagnosis result of the device fault is determined according to the search path.

[0012] Optionally, the extending and rewriting the problem description of the equipment failure to obtain an extended description of the equipment failure includes:

[0013] determining context information based on the problem description;

[0014] Determining, by means of a large language model, supplementary information of the problem description according to the context information;

[0015] The problem description is supplemented according to the supplementary information to obtain the extended description of the device failure.

[0016] Optionally, decomposing the extended description into a plurality of sub-questions includes:

[0017] Identifying the extended description to obtain a plurality of key words and phrases in the extended description;

[0018] Determining key components included in the key phrase according to the key phrase;

[0019] A problem description with the key component as the main body is generated according to each of the key components, and the problem description is used as the sub-problem corresponding to the key component.

[0020] Optionally, searching from the knowledge graph according to each of the sub-questions to obtain a retrieval sub-graph corresponding to each of the sub-questions includes:

[0021] Through the knowledge graph, entity linking is performed on the key components of each of the sub-problems to obtain specific nodes of the key components in the knowledge graph;

[0022] Retrieving multiple query nodes from the knowledge graph according to the specific node;

[0023] The retrieval subgraph is constructed according to the query node.

[0024] Optionally, the retrieving a plurality of query nodes from the knowledge graph according to the specific node includes:

[0025] Encoding the key component of each of the sub-problems through an encoder to obtain an entity vector corresponding to the key component;

[0026] Search the knowledge graph according to the entity vector to obtain the similarity between the specific node and other nodes;

[0027] According to the similarity, a preset number of the other nodes are selected as the query nodes.

[0028] Optionally, selecting a preset number of the other nodes as the query nodes according to the similarity includes:

[0029] According to the similarity, the other nodes are sorted by semantic similarity and topic consistency to obtain a node sequence;

[0030] Starting from the node with the highest similarity, the preset number of other nodes are selected in the node sequence in descending order as the query nodes.

[0031] Optionally, constructing the retrieval subgraph according to the query node includes:

[0032] The key component corresponding to each query node is taken as a central entity, and multiple one-hop relationships of the central entity are retrieved in the knowledge graph;

[0033] Convert the one-hop relationship into a string;

[0034] The character string, the key component corresponding to the character string and the sub-question are merged, and the merged result is input into the large language model for screening to obtain the retrieval subgraph corresponding to the sub-question.

[0035] Optionally, performing dynamic reasoning according to the retrieval subgraph to adjust the retrieval path of the retrieval subgraph in the knowledge graph includes:

[0036] Constructing a search path according to the search subgraph;

[0037] Converting the search subgraph into a subgraph sequence;

[0038] According to the relevant knowledge retrieved from the knowledge graph by the searched part of the search path, the unsearched part of the search path is adjusted through the subgraph sequence to obtain the adjusted search path of the search subgraph, and the search path of the search subgraph is continuously adjusted until the search is completed.

[0039] Optionally, determining the diagnosis result of the device fault according to the search path includes:

[0040] Searching in the knowledge graph according to the search path to obtain an answer to the sub-question;

[0041] The answers to all the sub-questions are summarized through a large language model to obtain the diagnosis result of the equipment failure.

[0042] In a second aspect, the present invention provides an equipment diagnosis system, comprising:

[0043] An expansion unit, used to expand and rewrite the problem description of the equipment failure to obtain an expanded description of the equipment failure;

[0044] A decomposition unit, used for decomposing the extended description into a plurality of sub-problems;

[0045] A subgraph construction unit, used to search from the knowledge graph according to each of the sub-questions to obtain a search subgraph corresponding to each of the sub-questions;

[0046] A retrieval path adjustment unit, used to perform dynamic reasoning based on the retrieval subgraph and adjust the retrieval path of the retrieval subgraph in the knowledge graph;

[0047] A summarizing unit is used to determine a diagnosis result of the device fault according to the retrieval path.

[0048] The equipment diagnosis method and system of the present invention firstly expand and rewrite the problem description of the equipment failure to obtain an expanded description of the equipment failure. By expanding and rewriting the problem description of the equipment failure, the context information and relevant terms of the problem are added, so that the problem description is more comprehensive and accurate. Then, the expanded description is decomposed to obtain multiple sub-problems, and the expanded description is subdivided into multiple sub-problems by using the problem decomposition technology, wherein each sub-problem is aimed at a specific aspect of the failure, and such subdivision makes the retrieval and reasoning more focused and in-depth. According to each sub-problem, the knowledge graph is retrieved to obtain the retrieval subgraph corresponding to each sub-problem, and the relevance of the retrieval information is ensured through entity recognition and linking. At the same time, since the knowledge of the knowledge graph can be updated regularly, the timeliness of the equipment fault diagnosis is ensured. Dynamic reasoning is performed according to the retrieval subgraph, and the retrieval path of the retrieval subgraph in the knowledge graph is adjusted; the diagnosis result of the equipment failure is determined according to the retrieval path, and the reasoning and correction are realized by dynamic reasoning and adjustment of the retrieval path, thereby overcoming the illusion and error propagation, and avoiding the problems of semantic inaccuracy and information redundancy. The knowledge graph can be flexibly explored, the most relevant information is screened out, the accuracy of the reasoning process is ensured, and the diagnosis accuracy is improved. Finally, the reasoning results of all sub-problems are combined to form a comprehensive diagnostic result. By utilizing the existing knowledge and relationships in the knowledge graph, the present invention not only improves the accuracy of diagnosis, but also improves the efficiency and adaptability of the diagnostic process, enabling it to cope with various complex and changeable equipment failure situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flowchart of a device diagnosis method in one embodiment of the present invention;

[0050] Figure 2 A schematic diagram of serialization of retrieving subgraphs in another embodiment of the present invention;

[0051] Figure 3 A schematic diagram of a flow chart of a device diagnosis method in another embodiment of the present invention;

[0052] Figure 4 It is a structural schematic diagram of a device diagnosis system in another embodiment of the present invention;

[0053] Figure 5The figure is a schematic structural diagram of an electronic device in another embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be interpreted as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are only for exemplary purposes and are not intended to limit the scope of protection of the present invention.

[0055] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0056] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to"; the term "based on" means "based at least in part on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiments". The relevant definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc. mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0057] It should be noted that the modifications of "one" and "plurality" mentioned in the present invention are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0058] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only used for illustrative purposes, and are not used to limit the scope of these messages or information.

[0059] In related technologies, once a device fails, it is necessary to quickly and accurately locate the cause of the failure and find a reasonable solution, as well as the risk of concurrent failures of related equipment. The traditional fault analysis process relies too much on the experience of engineers, with low utilization of fault data, high requirements for professional maintenance of equipment failures, easy loss of historical experience knowledge, and lack of scientific basis for optimization decisions. The analysis cycle is often long and prone to omissions. Inaccurate fault analysis will prolong the retention time of the fault and affect the normal operation of the equipment. The equipment maintenance method based on large language models mainly relies on text blocks to store data and cannot mine invisible associations. These text blocks are usually unstructured or semi-structured. With the expansion of knowledge, the addition of new data may require the reorganization of existing text blocks, or even repeated storage of multiple versions of similar content, resulting in low storage efficiency and difficulty in effective management. At the same time, the expansion of data scale will also significantly affect the retrieval performance of the system, and when faced with complex or semantically ambiguous queries, it is easily interfered by more noise data.

[0060] In view of the problems existing in the above-mentioned related technologies, this embodiment provides a device diagnosis method and system.

[0061] Combination Figure 1 As shown, an embodiment of the present invention provides a device diagnosis method, including:

[0062] The problem description of the equipment failure is expanded and rewritten to obtain an expanded description of the equipment failure.

[0063] Specifically, the original problem description of the equipment failure is extended and rewritten to obtain a more detailed extended description of the failure. The purpose of the extended rewriting is to enrich the original problem by adding relevant terms or concepts, thereby enhancing the effect of fault diagnosis retrieval query. In a preferred embodiment of the present invention, a large language model can be used to understand the context of the problem and generate a rewritten problem with more details. The rewritten problem will not only describe the fault phenomenon, but also include information such as device number, user operation, normal device performance, possible causes of the failure, solutions, and similar cases, thereby providing a more comprehensive context for subsequent sub-problem decomposition and knowledge graph retrieval.

[0064] The extended description is decomposed into multiple sub-problems.

[0065] Specifically, after obtaining the extended description, these descriptions are decomposed to form multiple sub-questions. Among them, the complex fault description is split into smaller and more specific queries through the problem decomposition technology, and each sub-query focuses on a specific aspect or component of the original query. By generating specific aspects or components separately, it helps to improve the accuracy of retrieval, so that each sub-question can retrieve relevant information from the knowledge graph in a targeted manner. In a preferred embodiment of the present invention, for the fault phenomenon of the equipment, the decomposed sub-questions may include inquiries about the normal operation phenomenon of the equipment, possible causes of the fault, solutions, etc.

[0066] According to each of the sub-questions, a search is performed from the knowledge graph to obtain a search sub-graph corresponding to each of the sub-questions.

[0067] Specifically, each sub-question is used to retrieve information from the knowledge graph, construct a corresponding retrieval sub-graph, map the entities in the sub-question to the corresponding entities in the knowledge graph, and retrieve information and relationships related to these entities. The retrieval sub-graph contains the entities and relationships most relevant to a specific sub-question, forming a local knowledge structure that is closely related to the query, and utilizing the existing knowledge and relationships in the knowledge graph, and mining potential information related to fault diagnosis through entity linking and sub-graph retrieval technology.

[0068] Dynamic reasoning is performed based on the retrieval subgraph to adjust the retrieval path of the retrieval subgraph in the knowledge graph.

[0069] Specifically, after the retrieval subgraph is constructed, it is necessary to perform dynamic reasoning based on these subgraphs and adjust the retrieval path. According to the designed fault handling process, the retrieval path of the retrieval subgraph in the knowledge graph is dynamically adjusted. A large language model is used to perform semantic analysis on the relationships in the retrieval subgraph, screen out the relationships most relevant to the sub-problem, and adjust the retrieval path to ensure the accuracy and relevance of the reasoning.

[0070] A diagnosis result of the device fault is determined according to the search path.

[0071] Specifically, the diagnostic results of the equipment fault are finally determined based on the adjusted search path, and the answers to all sub-questions are combined to form a comprehensive fault diagnosis report. This method can provide more accurate and comprehensive fault diagnosis, reduce reliance on past experience, and improve analysis efficiency and accuracy.

[0072] The equipment diagnosis method of the present invention firstly expands and rewrites the problem description of the equipment failure to obtain an expanded description of the equipment failure. By expanding and rewriting the problem description of the equipment failure, the context information and relevant terms of the problem are added, so that the problem description is more comprehensive and accurate. Then, the expanded description is decomposed to obtain multiple sub-problems, and the expanded description is subdivided into multiple sub-problems by using the problem decomposition technology, wherein each sub-problem is aimed at a specific aspect of the failure, and such subdivision makes the retrieval and reasoning more focused and in-depth. According to each sub-problem, a search is performed from the knowledge graph to obtain a search sub-graph corresponding to each sub-problem, and the relevance of the search information is ensured through entity recognition and linking. At the same time, since the knowledge of the knowledge graph can be updated regularly, the timeliness of the equipment fault diagnosis is ensured. Dynamic reasoning is performed according to the search sub-graph, and the search path of the search sub-graph in the knowledge graph is adjusted; the diagnosis result of the equipment failure is determined according to the search path, and the reasoning and correction are realized by dynamic reasoning and adjustment of the search path, thereby overcoming the illusion and error propagation, and avoiding the problems of semantic inaccuracy and information redundancy. The knowledge graph can be flexibly explored, the most relevant information is screened out, the accuracy of the reasoning process is ensured, and the diagnosis accuracy is improved. Finally, the reasoning results of all sub-problems are combined to form a comprehensive diagnostic result. By utilizing the existing knowledge and relationships in the knowledge graph, the present invention not only improves the accuracy of diagnosis, but also improves the efficiency and adaptability of the diagnostic process, enabling it to cope with various complex and changeable equipment failure situations.

[0073] Optionally, the extending and rewriting the problem description of the equipment failure to obtain an extended description of the equipment failure includes:

[0074] determining context information based on the problem description;

[0075] Determining, by means of a large language model, supplementary information of the problem description according to the context information;

[0076] The problem description is supplemented according to the supplementary information to obtain the extended description of the device failure.

[0077] Specifically, the problem description is usually short and has limited information content. The purpose of problem description expansion is to improve the diagnostic answer effect by using additional relevant terms or concepts to supplement or improve the original query explicitness, make the answer content more comprehensive, summarize the normal equipment maintenance thinking process, and generate more comprehensive and appropriate questions to enhance the fault diagnosis retrieval query inspection effect. In general, the process of expanding and rewriting the problem description of equipment failure first needs to determine the context information of the problem description, which includes identifying key elements such as fault phenomenon, equipment model, and user operation. For example, equipment failure phenomena include motor overheating and spindle rotation weakness. Then, using a large language model, based on this contextual information, the model can understand the semantics of the problem and determine the information that needs to be supplemented, such as possible causes of failure, related maintenance cases, and performance during normal operation. Finally, combined with this supplementary information, the original problem description is expanded to form a more detailed extended description of the equipment failure. This process enables the problem description to include not only the fault phenomenon, but also other key information required in the fault analysis and resolution process, providing a solid foundation for subsequent sub-problem decomposition and knowledge graph retrieval.

[0078] For example, the problem description of equipment failure "After the CQ8834 equipment was turned on, the motor overheated during operation, the spindle could not rotate, and there was a lot of vibration and noise. How to solve it? What are the hidden dangers of related equipment?" is expanded to: "The equipment number is CQ8834. The user turned on the equipment. What are the phenomena exhibited by a normal equipment? The fault phenomenon is: the motor overheated during operation, the spindle could not rotate, and there was a lot of vibration and noise. The main fault location is the motor. What are the possible related fault phenomena? What is the most likely cause of the fault? What are the main solutions? What are similar maintenance cases?" etc.

[0079] In an embodiment of the present invention, by extending the rewriting step, the accuracy and efficiency of equipment fault diagnosis are significantly improved, making the problem more comprehensive and covering multiple aspects of fault diagnosis, thereby making subsequent sub-problems more precise, the retrieval subgraph more relevant, and the dynamic reasoning more in-depth.

[0080] Optionally, decomposing the extended description into a plurality of sub-questions includes:

[0081] Identifying the extended description to obtain a plurality of key words and phrases in the extended description;

[0082] Determining key components included in the key phrase according to the key phrase;

[0083] A problem description with the key component as the main body is generated according to each of the key components, and the problem description is used as the sub-problem corresponding to the key component.

[0084] Specifically, the problem decomposition technique decomposes the extended fault description problem into smaller and more specific sub-queries, so that the query can become more accurate. Each sub-query usually focuses on a specific aspect or component of the original query, each clause represents a different relationship, and retrieves the associated triples of each clause in the knowledge graph in turn. The process of decomposing the extended description of the equipment fault to obtain multiple sub-questions first involves the detailed identification of the extended description and extracting multiple key words from it.

[0085] For example, the extended description "The device number is CQ8834, the user has powered on the device. What are the normal phenomena of the device? The fault phenomenon is: the motor overheats during operation, the spindle cannot rotate, accompanied by large vibration and noise. The main fault part is the motor. What are the possible related fault phenomena? What is the most likely cause of the fault? What are the main solutions? What are similar maintenance cases?" is decomposed into sub-question 1: "The device number is CQ8834, the user has powered on the device. The fault phenomenon is: the motor overheats during operation, the spindle cannot rotate, accompanied by large vibration and noise. The main fault part is the motor. What is the normal operation phenomenon?", Sub-question 2: "The device number is CQ8834, the user has powered on the device. The fault phenomenon is: the motor overheats during operation, the spindle cannot rotate, accompanied by large vibration and noise. The main fault part is the motor. What is the normal operation phenomenon?" , the spindle cannot rotate easily, accompanied by large vibration and noise, the main fault part is the motor, what are the related fault phenomena? ", Sub-question 3: "The equipment number is CQ8834, the user has powered on the equipment, the fault phenomenon is: the motor overheats during operation, the spindle cannot rotate easily, accompanied by large vibration and noise, the main fault part is the motor, the normal operation phenomenon is state A, the related fault phenomenon is state B, the possible cause of the fault is? ", Sub-question 4: "The equipment number is CQ8834, the user has powered on the equipment, the fault phenomenon is: the motor overheats during operation, the spindle cannot rotate easily, accompanied by large vibration and noise, the main fault part is the motor, the normal operation phenomenon is state A, the related fault phenomenon is state B, the cause of the fault is cause C, what is the solution" etc.

[0086] In the embodiment of the present invention, the comprehensiveness and systematicness of the fault diagnosis process are ensured by decomposing the extended description into multiple sub-questions, each of which targets a specific fault aspect, making the information retrieved from the knowledge graph more accurate and relevant. Secondly, this method improves the efficiency of diagnosis because each sub-question can be processed independently, so that multiple queries can be processed in parallel, speeding up the entire diagnosis process.

[0087] Optionally, searching from the knowledge graph according to each of the sub-questions to obtain a retrieval sub-graph corresponding to each of the sub-questions includes:

[0088] Through the knowledge graph, entity linking is performed on the key components of each of the sub-problems to obtain specific nodes of the key components in the knowledge graph;

[0089] Retrieving multiple query nodes from the knowledge graph according to the specific node;

[0090] The retrieval subgraph is constructed according to the query node.

[0091] Specifically, the process of retrieving from the knowledge graph according to each sub-question first involves entity linking of the key components in the sub-question. If the sub-question is what the normal operation phenomenon of the equipment is, the equipment is used as a key component, and the corresponding specific node is found in the knowledge graph through the entity linking technology, that is, the string of the equipment is mapped to the corresponding entity in the knowledge graph to obtain the specific node. Then, based on this node, the system retrieves multiple query nodes related to the equipment from the knowledge graph, which may include the technical parameters of the equipment, historical fault records, maintenance cases, etc. Then, based on these query nodes, a retrieval subgraph is constructed, which contains entities and relationships related to the equipment, providing rich contextual information for further dynamic reasoning. This process ensures that each sub-question can retrieve information directly related to it from the knowledge graph, providing accurate data support for subsequent fault diagnosis.

[0092] In the embodiments of the present invention, the knowledge graph supports the accumulation and updating of knowledge, because the knowledge graph can be regularly updated to reflect the latest equipment data and maintenance experience. The accuracy of fault diagnosis is improved, because the retrieval subgraph contains entities and relationships directly related to the sub-problem, which helps the system understand the context of the fault more accurately. Secondly, the pertinence and efficiency of the diagnosis process are also enhanced, because each sub-problem can quickly locate the most relevant information, reducing the interference of irrelevant information.

[0093] Optionally, the retrieving a plurality of query nodes from the knowledge graph according to the specific node includes:

[0094] Encoding the key component of each of the sub-problems through an encoder to obtain an entity vector corresponding to the key component;

[0095] Search the knowledge graph according to the entity vector to obtain the similarity between the specific node and other nodes;

[0096] According to the similarity, a preset number of the other nodes are selected as the query nodes.

[0097] Specifically, the process of retrieving multiple query nodes from the knowledge graph begins with encoding the key components in each sub-question using an encoder. In a preferred embodiment of the present invention, if the sub-question involves a device number of CQ8834, then the key component CQ8834 will be converted into an entity vector by the encoder. This entity vector represents the semantic and attribute information of the device number. Subsequently, the system uses this entity vector to search in the knowledge graph and calculates its similarity with other nodes in the knowledge graph. This step involves complex vector space models and similarity calculation methods, such as cosine similarity. Based on these similarity scores, the system selects a preset number of the most similar nodes as query nodes, which may represent other equipment, fault cases, maintenance records, etc. related to the CQ8834 device. This process ensures the relevance and diversity of the retrieval results, and provides a rich source of information for constructing retrieval subgraphs and subsequent dynamic reasoning.

[0098] In the embodiment of the present invention, the accuracy and relevance of the search are improved by using a search method based on entity vectors and similarity, because the search results are screened based on the similarity between the entity vectors and the nodes in the knowledge graph, which helps to find the most relevant information for the sub-question. Secondly, the adaptability of the system is enhanced, and it can handle a variety of different sub-questions, so that it does not rely on static keyword matching, but relies on more flexible and comprehensive semantic similarity calculation.

[0099] Optionally, selecting a preset number of the other nodes as the query nodes according to the similarity includes:

[0100] According to the similarity, the other nodes are sorted by semantic similarity and topic consistency to obtain a node sequence;

[0101] Starting from the node with the highest similarity, the preset number of other nodes are selected in the node sequence in descending order as the query nodes.

[0102] Specifically, a large language model can be used to write consistent hints to reorder the topic consistency of the entity background (the entity's subgraph retrieval results, i.e., other nodes) between the entity and the candidate entity, so that the similarity and consistency issues are comprehensively considered when linking entities. The consistency here refers to the local relevance, i.e., the topic similarity between the fault description text containing the extracted entity and the background text corresponding to the candidate entity. First, the key components in each sub-problem are converted into entity vectors through the encoder. This conversion is based on the attributes and context information of the entity, mapping the textual entity to a vector representation in a high-dimensional space. Subsequently, nodes similar to these entity vectors are retrieved in the knowledge graph and their similarity is calculated, which usually involves metrics such as cosine similarity. Next, the similarity sorting sorts all candidate nodes from high to low according to the similarity to form a node sequence. Finally, a preset number of nodes with the highest similarity are selected from this sequence as query nodes. These nodes will be used to construct a retrieval subgraph for each sub-problem, providing an accurate data basis for dynamic reasoning.

[0103] In the embodiment of the present invention, by ensuring that the selected nodes are highly relevant to the query entity, the accuracy and efficiency of the retrieval are improved. Secondly, by sorting and selecting the nodes with the highest similarity, the system can effectively focus on the areas most likely to contain useful information, reducing the impact of noise data.

[0104] Optionally, constructing the retrieval subgraph according to the query node includes:

[0105] The key component corresponding to each query node is taken as a central entity, and multiple one-hop relationships of the central entity are retrieved in the knowledge graph;

[0106] Convert the one-hop relationship into a string;

[0107] The character string, the key component corresponding to the character string and the sub-question are merged, and the merged result is input into the large language model for screening to obtain the retrieval subgraph corresponding to the sub-question.

[0108] Specifically, the key components corresponding to the query nodes are regarded as central entities, and multiple one-hop relations of these central entities, i.e., entities and relations directly connected to the central entities, are retrieved in the knowledge graph. In this process, a graph traversal is performed in the knowledge graph, starting from the central entity, exploring its direct neighbor nodes and the edges connecting them, so as to obtain information directly related to the central entity. Next, the one-hop relations are converted into strings for further processing. These strings are merged with their corresponding key components and sub-problems to form a new input, and then this merged input is fed into the large language model. Based on this input, the large language model understands the semantic connection between the one-hop relations and the sub-problems, evaluates the relevant information of these relations, and accordingly filters and constructs the retrieval subgraph directly related to the sub-problems, ensuring that the retrieval subgraph contains the most valuable information for solving specific sub-problems, thereby improving the pertinence and accuracy of fault diagnosis. Finally, each entity and the preset number of selected nodes constitute the associated retrieval subgraph. According to the designed template, the associated knowledge graph of the query can be represented as a list of triples, which are converted into natural language paragraphs through the template to better retrieve relevant knowledge.

[0109] In a preferred embodiment of the present invention, one-hop relationships of the central entity are explored. Since not all one-hop relationships are relevant to the question, the obtained relationship set needs to be pruned. First, the relationship is linearly converted into a string and its corresponding entity and question set are combined together to form a prompt, which is then input into a large language model to filter out multiple most relevant relationships. Finally, a preset number of relationships are selected from the subsequent relationships of each entity.

[0110] In an embodiment of the present invention, a retrieval subgraph is constructed based on a one-hop relationship so that the retrieval subgraph can accurately reflect the core requirements of the sub-problem, and the one-hop relationship is converted into a string and merged with the sub-problem, which provides rich contextual information for the large language model, enabling it to perform in-depth semantic analysis and relevance evaluation, thereby improving the quality of the retrieval subgraph.

[0111] Optionally, performing dynamic reasoning according to the retrieval subgraph to adjust the retrieval path of the retrieval subgraph in the knowledge graph includes:

[0112] Constructing a search path according to the search subgraph;

[0113] Converting the search subgraph into a subgraph sequence;

[0114] According to the relevant knowledge retrieved from the knowledge graph by the searched part of the search path, the unsearched part of the search path is adjusted through the subgraph sequence to obtain the adjusted search path of the search subgraph, and the search path of the search subgraph is continuously adjusted until the search is completed.

[0115] Specifically, the process of dynamic reasoning and retrieval path adjustment is a key link in data processing. First, the system constructs an initial retrieval path based on the retrieval subgraph, which involves identifying the key entities and relationships in the subgraph and determining the logical order between them. Next, the associated retrieval subgraph of the retrieval subgraph is converted into a subgraph sequence, and the structural information of the subgraph is linearized to facilitate subsequent processing and analysis. Then, the system dynamically adjusts the unretrieved part of the retrieval path through the subgraph sequence based on the relevant knowledge retrieved from the knowledge graph of the retrieved part, wherein the retrieval path is updated according to the newly retrieved information, new retrieval nodes are added, or existing retrieval nodes are reordered to optimize the retrieval efficiency and the relevance of the results. This dynamic adjustment continues until the retrieval of the entire retrieval path is completed, ensuring that the retrieval path can flexibly adapt to changes in information in the knowledge graph and new discoveries in the sub-problem solving process.

[0116] In a preferred embodiment of the present invention, a small number of sample prompts can be selected to guide the large language model to generate reasoning rules and specific task actions based on the processing logic of fault diagnosis, and at the same time obtain external information through interaction, and integrate the external information into the entire dynamic reasoning process to complete the diagnosis of the fault, including but not limited to the association of normal phenomena and fault phenomena. Fault causes, fault solutions, concurrent fault phenomena, similar fault cases, detection tools, etc., through interaction with the retrieval subgraph, the retrieved subgraph is converted into a sequence representation to verify and modify the entire maintenance reasoning trace, and the reasoning and correction are performed while overcoming illusions and error propagation. Based on the relevant knowledge retrieved from the retrieval subgraph at each step, reasoning rules and specific task actions are generated for the large language model based on the processing logic of fault diagnosis, and a reasoning plan is generated through interaction.

[0117] In a preferred embodiment of the present invention, the associated search subgraph of the entity search in the search subgraph is converted into a sequence, wherein the processing method for converting into a sequence is as follows: the central node is taken as the root node, converted into a tree structure, and then a layer-by-layer traversal is performed, and the leaf nodes selected at each layer are fixed. If they are not enough, the sequence is completed with PAD characters. The serialization result is as follows: Figure 2 shown.

[0118] In the embodiments of the present invention, the flexibility and adaptability of retrieval are improved by the method of dynamic reasoning and retrieval path adjustment, because the retrieval path can be adjusted according to the real-time information retrieved, so as to more accurately capture the knowledge most relevant to the sub-question. Secondly, by allowing the system to continuously learn and adapt during the retrieval process, the retrieval strategy is optimized, and the reasoning ability of the system is enhanced.

[0119] Optionally, determining the diagnosis result of the device fault according to the search path includes:

[0120] Searching in the knowledge graph according to the search path to obtain an answer to the sub-question;

[0121] The answers to all the sub-questions are summarized through a large language model to obtain the diagnosis result of the equipment failure.

[0122] Specifically, the system performs precise retrieval in the knowledge graph according to the constructed and adjusted retrieval path, obtains answers for each sub-question, and utilizes the structured knowledge in the knowledge graph and the relevant information in the retrieval sub-graph to ensure that the answer to each sub-question is based on the most relevant and accurate data. Then, these scattered sub-question answers are input into the large language model, which integrates and summarizes the answers to all sub-questions through its advanced natural language processing capabilities to form a comprehensive equipment fault diagnosis result. This summarization process involves not only simple answer aggregation, but also analysis and reasoning of the relationship between answers to ensure that the final diagnosis result is coherent and logical. In a preferred embodiment of the present invention, a large language model is used to summarize multiple rounds of answers based on questions and answers. At the same time, it is worth mentioning that when generating questions in each round, while inputting external knowledge and questions, the large language model is required to generate answers based entirely on the provided external knowledge without using its own knowledge.

[0123] In an embodiment of the present invention, by retrieving answers to sub-questions in a knowledge graph, the system can ensure that each answer is based on the latest and most relevant data, thereby improving the accuracy of the diagnosis.

[0124] In summary, combined with Figure 3 As shown in the figure, the fault description is first expanded, and then the expanded description is broken down into multiple sub-questions, each of which targets a specific fault feature or doubt. Next, these sub-questions are used to plan specific fault diagnosis logic to lay the foundation for subsequent diagnosis work. In the diagnostic logic planning stage, the system matches the entities in the text with the corresponding nodes in the knowledge graph through entity recognition and linking technology, and sorts them according to relevance. This process enables the system to accurately locate the specific components and concepts related to the fault. Subsequently, the system conducts in-depth retrieval in the knowledge graph, converts the retrieved subgraphs into an easy-to-process serialized form, and expresses this information in natural language. Based on this information, all retrieved information is integrated and summarized to output a comprehensive fault diagnosis result.

[0125] Combination Figure 4 As shown, the present invention also provides an equipment diagnosis system, comprising:

[0126] An expansion unit, used to expand and rewrite the problem description of the equipment failure to obtain an expanded description of the equipment failure;

[0127] A decomposition unit, used for decomposing the extended description into a plurality of sub-problems;

[0128] A subgraph construction unit, used to search from the knowledge graph according to each of the sub-questions to obtain a search subgraph corresponding to each of the sub-questions;

[0129] A retrieval path adjustment unit, used to perform dynamic reasoning based on the retrieval subgraph and adjust the retrieval path of the retrieval subgraph in the knowledge graph;

[0130] A summarizing unit is used to determine a diagnosis result of the device fault according to the retrieval path.

[0131] The extension unit is further used to determine context information based on the problem description; determine supplementary information of the problem description based on the context information through a large language model; and supplement the problem description based on the supplementary information to obtain the extended description of the device failure.

[0132] Optionally, the decomposition unit is further used to identify the extended description to obtain multiple key words and phrases in the extended description; determine key components contained in the key words and phrases based on the key words and phrases; generate a problem description with the key component as the main body based on each key component, and use the problem description as the sub-problem corresponding to the key component.

[0133] Optionally, the subgraph construction unit is also used to perform entity linking on the key components of each of the sub-problems through the knowledge graph to obtain specific nodes of the key components in the knowledge graph; retrieve multiple query nodes from the knowledge graph based on the specific nodes; and construct the retrieval subgraph based on the query nodes.

[0134] Optionally, the subgraph construction unit is further used to encode the key components of each of the sub-problems through an encoder to obtain entity vectors corresponding to the key components; search the knowledge graph according to the entity vectors to obtain the similarity between the specific node and other nodes; and select a preset number of the other nodes as the query nodes based on the similarity.

[0135] Optionally, the subgraph construction unit is further used to sort the other nodes according to the size of the similarity to obtain a node sequence; sort the other nodes according to the similarity by semantic similarity and topic consistency to obtain a node sequence; starting from the highest similarity, select the preset number of other nodes in the node sequence in descending order as the query nodes.

[0136] Optionally, the subgraph construction unit is further used to take the key component corresponding to each of the query nodes as a central entity, and retrieve multiple one-hop relationships of the central entity in the knowledge graph; convert the one-hop relationship into a string; merge the string and the key component corresponding to the string with the sub-question, and then input the merged result into the large language model for screening to obtain the retrieval subgraph corresponding to the sub-question.

[0137] Optionally, the retrieval path adjustment unit is also used to construct a retrieval path based on the retrieval subgraph; convert the retrieval subgraph into a subgraph sequence; adjust the unretrieved part of the retrieval path through the subgraph sequence according to the relevant knowledge retrieved in the knowledge graph based on the retrieved part of the retrieval path to obtain the adjusted retrieval path of the retrieval subgraph, and continue to adjust the retrieval path of the retrieval subgraph until the retrieval is completed.

[0138] Optionally, the decomposition unit is further used to search in the knowledge graph according to the retrieval path to obtain answers to the sub-questions; and summarize the answers to all the sub-questions through a large language model to obtain the diagnosis result of the equipment failure.

[0139] The equipment diagnosis system of the present invention firstly expands and rewrites the problem description of the equipment failure to obtain an expanded description of the equipment failure. By expanding and rewriting the problem description of the equipment failure, the context information and relevant terms of the problem are added, so that the problem description is more comprehensive and accurate. Then, the expanded description is decomposed to obtain multiple sub-problems, and the expanded description is subdivided into multiple sub-problems by using the problem decomposition technology, wherein each sub-problem is aimed at a specific aspect of the failure, and such subdivision makes the retrieval and reasoning more focused and in-depth. According to each sub-problem, the knowledge graph is retrieved to obtain the retrieval subgraph corresponding to each sub-problem, and the relevance of the retrieval information is ensured through entity recognition and linking. At the same time, since the knowledge of the knowledge graph can be updated regularly, the timeliness of the equipment fault diagnosis is guaranteed. Dynamic reasoning is performed according to the retrieval subgraph, and the retrieval path of the retrieval subgraph in the knowledge graph is adjusted; the diagnosis result of the equipment failure is determined according to the retrieval path, and the reasoning and correction are realized by dynamic reasoning and adjusting the retrieval path, thereby overcoming the illusion and error propagation, and avoiding the problems of semantic inaccuracy and information redundancy. The knowledge graph can be flexibly explored, the most relevant information is screened out, the accuracy of the reasoning process is ensured, and the diagnosis accuracy is improved. Finally, the reasoning results of all sub-problems are combined to form a comprehensive diagnostic result. By utilizing the existing knowledge and relationships in the knowledge graph, the present invention not only improves the accuracy of diagnosis, but also improves the efficiency and adaptability of the diagnostic process, enabling it to cope with various complex and changeable equipment failure situations.

[0140] Combination Figure 5As shown, an electronic device 500 provided by an embodiment of the present invention includes a memory 510 and a processor 520; the memory 510 is used to store a computer program; the processor 520 is used to implement the device diagnosis method as described above when executing the computer program.

[0141] In other words, an electronic device 500 includes a memory 510 and a processor 520 coupled to the memory 510; the memory 510 is configured to store a computer program; and the processor 520 is configured to perform the following operations when executing the computer program:

[0142] Expand and rewrite the problem description of the equipment failure to obtain an expanded description of the equipment failure;

[0143] Decomposing the extended description into a plurality of sub-problems;

[0144] Searching from the knowledge graph according to each of the sub-questions to obtain a search sub-graph corresponding to each of the sub-questions;

[0145] Performing dynamic reasoning based on the retrieval subgraph to adjust the retrieval path of the retrieval subgraph in the knowledge graph;

[0146] A diagnosis result of the device fault is determined according to the search path.

[0147] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the device diagnosis method as described above is implemented.

[0148] In other words, a non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor performs the following operations:

[0149] Expand and rewrite the problem description of the equipment failure to obtain an expanded description of the equipment failure;

[0150] Decomposing the extended description into a plurality of sub-problems;

[0151] Searching from the knowledge graph according to each of the sub-questions to obtain a search sub-graph corresponding to each of the sub-questions;

[0152] Performing dynamic reasoning based on the retrieval subgraph to adjust the retrieval path of the retrieval subgraph in the knowledge graph;

[0153] A diagnosis result of the device fault is determined according to the search path.

[0154] An electronic device 500 that can be used as a server or client of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device 500 is intended to represent various forms of digital electronic computer equipment, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 500 can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples, and are not intended to limit the implementation of the present invention described and / or required herein.

[0155] The electronic device 500 includes a computing unit, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for the operation of the device can also be stored. The computing unit, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0156] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc. In the present application, the unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment of the present invention. In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0157] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A device diagnosis method, characterized in that: include: Expand and rewrite the problem description of the equipment failure to obtain an expanded description of the equipment failure; Decomposing the extended description into a plurality of sub-problems; Searching from the knowledge graph according to each of the sub-questions to obtain a search sub-graph corresponding to each of the sub-questions; Performing dynamic reasoning based on the retrieval subgraph to adjust the retrieval path of the retrieval subgraph in the knowledge graph; A diagnosis result of the device fault is determined according to the search path.

2. The device diagnosis method according to claim 1, characterized in that: The problem description of the equipment failure is extended and rewritten to obtain an extended description of the equipment failure, including: determining context information based on the problem description; Determining, by means of a large language model, supplementary information of the problem description according to the context information; The problem description is supplemented according to the supplementary information to obtain the extended description of the device failure.

3. The device diagnosis method according to claim 1, characterized in that: The extended description is decomposed into multiple sub-problems, including: Identifying the extended description to obtain a plurality of key words and phrases in the extended description; Determining key components included in the key phrase according to the key phrase; A problem description with the key component as the main body is generated according to each of the key components, and the problem description is used as the sub-problem corresponding to the key component.

4. The device diagnosis method according to claim 3, characterized in that: The step of searching from the knowledge graph according to each of the sub-questions to obtain a search sub-graph corresponding to each of the sub-questions includes: Through the knowledge graph, entity linking is performed on the key components of each of the sub-problems to obtain specific nodes of the key components in the knowledge graph; Retrieving multiple query nodes from the knowledge graph according to the specific node; The retrieval subgraph is constructed according to the query node.

5. The device diagnosis method according to claim 4, characterized in that: The retrieving a plurality of query nodes from the knowledge graph according to the specific node includes: Encoding the key component of each of the sub-problems through an encoder to obtain an entity vector corresponding to the key component; Search the knowledge graph according to the entity vector to obtain the similarity between the specific node and other nodes; According to the similarity, a preset number of the other nodes are selected as the query nodes.

6. The device diagnosis method according to claim 5, characterized in that: The selecting a preset number of the other nodes as the query nodes according to the similarity includes: According to the similarity, the other nodes are sorted by semantic similarity and topic consistency to obtain a node sequence; Starting from the node with the highest similarity, the preset number of other nodes are selected in the node sequence in descending order as the query nodes.

7. The device diagnosis method according to claim 4, characterized in that: The step of constructing the retrieval subgraph according to the query node includes: The key component corresponding to each query node is taken as a central entity, and multiple one-hop relationships of the central entity are retrieved in the knowledge graph; Convert the one-hop relationship into a string; The character string, the key component corresponding to the character string and the sub-question are merged, and the merged result is input into the large language model for screening to obtain the retrieval subgraph corresponding to the sub-question.

8. The device diagnosis method according to claim 1, characterized in that: The step of performing dynamic reasoning according to the search subgraph and adjusting the search path of the search subgraph in the knowledge graph includes: Constructing a search path according to the search subgraph; Converting the search subgraph into a subgraph sequence; According to the relevant knowledge retrieved from the knowledge graph by the searched part of the search path, the unsearched part of the search path is adjusted through the subgraph sequence to obtain the adjusted search path of the search subgraph, and the search path of the search subgraph is continuously adjusted until the search is completed.

9. The device diagnosis method according to claim 1, characterized in that: Determining the diagnosis result of the device fault according to the search path includes: Searching in the knowledge graph according to the search path to obtain an answer to the sub-question; The answers to all the sub-questions are summarized through a large language model to obtain the diagnosis result of the equipment failure.

10. A device diagnostic system, characterized in that: include: An expansion unit, used to expand and rewrite the problem description of the equipment failure to obtain an expanded description of the equipment failure; A decomposition unit, used for decomposing the extended description into a plurality of sub-problems; A subgraph construction unit, used to search from the knowledge graph according to each of the sub-questions to obtain a search subgraph corresponding to each of the sub-questions; A retrieval path adjustment unit, used to perform dynamic reasoning based on the retrieval subgraph and adjust the retrieval path of the retrieval subgraph in the knowledge graph; A summarizing unit is used to determine a diagnosis result of the device fault according to the retrieval path.

Images