Fault maintenance method and device, equipment, storage medium and computer program product

Abstract

The invention discloses a fault maintenance method and device, equipment, a storage medium and a computer program product, and relates to the technical field of computers, and the fault maintenance method comprises the steps that when a query problem of a target equipment fault reason is received, intention recognition is conducted on the query problem through a target language large model, and an intention recognition result is obtained; calling a vocabulary resource library according to the intention recognition result, and querying a fault maintenance knowledge graph generated based on a target document large model according to a target vocabulary obtained by calling to obtain a plurality of recall answers; performing network resource retrieval according to the recall answer to obtain a network retrieval result of fault classification; and determining a fault classification according to the recall answer and the network retrieval result, and generating a fault solution recommendation scheme according to the fault classification. Through the mode, customized and accurate fault removal suggestions are provided, and meanwhile, the efficiency and the accuracy of fault maintenance are improved by relying on the knowledge graph and the large model.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a fault repair method, apparatus, device, storage medium and computer program product. Background Art

[0002] The maintenance of various types of high-end mechanical equipment, vehicles, and technological products is a highly professional and complex task, which requires not only maintenance personnel to have solid theoretical knowledge and practical experience, but also to be able to quickly and accurately locate the cause of the failure and take effective measures. The traditional maintenance model relies on manual review of a large amount of technical information and past cases, which is a time-consuming and error-prone process.

[0003] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0004] The main purpose of the present application is to provide a fault repair method, apparatus, device, storage medium and computer program product, aiming to solve the technical problem that equipment fault troubleshooting is time-consuming and error-prone.

[0005] To achieve the above objectives, the present application proposes a fault repair method, which comprises:

[0006] When receiving a query question about the cause of a target device failure, performing intent recognition on the query question through the target language large model to obtain an intent recognition result;

[0007] Calling a vocabulary resource library according to the intention recognition result, and querying a fault maintenance knowledge graph generated based on a target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers;

[0008] Perform network resource search according to the recalled answer to obtain network search results of fault classification;

[0009] A fault classification is determined according to the recall answer and the network search result, and a fault solution recommendation is generated according to the fault classification.

[0010] Optionally, before the step of calling the vocabulary resource library according to the intention recognition result and querying the fault maintenance knowledge graph generated based on the target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers, the step further includes:

[0011] Divide the content of the preset maintenance manual into multiple long text segments, input the long text segments into the target document macro model for segmentation processing, and obtain the maintenance manual content according to the processing results;

[0012] Generate a historical fault knowledge base based on historical fault data;

[0013] A fault maintenance knowledge graph is generated based on the historical fault knowledge base and the maintenance manual content.

[0014] Optionally, the step of dividing the content of the preset maintenance manual into a plurality of long text segments includes:

[0015] Obtaining a text file of a preset maintenance manual, and extracting the content of the target manual according to the text file, wherein the content extraction includes paragraph, picture, table and page number extraction;

[0016] Perform label extraction and summary extraction according to the target manual content to obtain the analysis manual content;

[0017] A plurality of long text segments are obtained based on the analysis of the manual content.

[0018] The step of obtaining a plurality of long text segments according to the content of the analysis manual comprises:

[0019] Determining manual semantic information based on analyzing the manual content;

[0020] The analyzed manual content is divided into a plurality of long text segments according to the manual semantic information.

[0021] Optionally, the step of generating a historical fault knowledge base according to historical fault data includes:

[0022] Converting the historical fault data into a question-answer pair format to obtain question-answer pair data;

[0023] Generate a historical fault file based on the question and answer data;

[0024] A historical fault knowledge base is generated according to the historical fault files.

[0025] Optionally, the step of generating a fault maintenance knowledge graph according to the historical fault knowledge base and the maintenance manual content includes:

[0026] Obtain maintenance keywords according to the maintenance manual content;

[0027] generating a keyword table according to the maintenance keywords;

[0028] The keyword table and the historical fault knowledge base are imported into the initial knowledge graph to obtain a fault maintenance knowledge graph.

[0029] Optionally, when receiving a query question about the cause of a target device failure, performing intent recognition on the query question through a target language large model to obtain an intent recognition result includes:

[0030] When receiving a query question about the cause of a target device failure, importing the query question into the target language macro model to obtain the user intention;

[0031] It is determined whether to call a vocabulary resource library according to the user intention, and an intention recognition result is obtained according to the determination result.

[0032] Optionally, when receiving a query question about the cause of a target device failure, the step of importing the query question into the target language macro model to obtain the user intention includes:

[0033] When receiving a query question about the cause of a target device failure, determining a query prompt word according to the query question;

[0034] Extracting subject name information from the query prompt word;

[0035] The subject name information and the query question are imported into the target language macro model to obtain the user intention.

[0036] Optionally, the step of calling a vocabulary resource library according to the intention recognition result, and querying a fault maintenance knowledge graph generated based on a target document large model according to the target vocabulary obtained by the call to obtain a plurality of recall answers includes:

[0037] Calling a vocabulary resource library according to the intention recognition result to obtain a plurality of target words corresponding to the query question;

[0038] According to the target vocabulary, a fault maintenance knowledge graph is searched to obtain a plurality of vocabulary-related questions;

[0039] A plurality of recalled answers are obtained according to the vocabulary-related questions.

[0040] Optionally, the step of performing a network resource search according to the recalled answer to obtain a network search result of the fault classification includes:

[0041] Screening the recalled answers through the target proxy node to obtain a screening result;

[0042] Determine the troubleshooting chapter number corresponding to the preset maintenance manual according to the screening result;

[0043] The target network resource is searched according to the troubleshooting chapter number to obtain a network search result.

[0044] In addition, to achieve the above-mentioned purpose, the present application also proposes a fault repair device, the fault repair device comprising:

[0045] An intent recognition module is used to, when receiving a query question about the cause of a target device failure, perform intent recognition on the query question through a target language large model to obtain an intent recognition result;

[0046] An answer recall module is used to call a vocabulary resource library according to the intention recognition result, and query the fault maintenance knowledge graph generated based on the target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers;

[0047] A network search module, used to search network resources according to the recalled answers to obtain network search results for fault classification;

[0048] A solution generation module is used to determine the fault classification according to the recalled answer and the network search result, and to generate a recommended fault solution according to the fault classification.

[0049] Optionally, the answer recall module is also used to divide the content of a preset maintenance manual into multiple long text segments, and input the long text segments into the target document model for segmentation processing, and obtain the maintenance manual content according to the processing results; generate a historical fault knowledge base according to historical fault data; and generate a fault maintenance knowledge graph according to the historical fault knowledge base and the maintenance manual content.

[0050] Optionally, the answer recall module is also used to obtain a text file of a preset maintenance manual, and perform content extraction based on the text file to obtain the target manual content, wherein the content extraction includes paragraph, picture, table and page number extraction; perform label extraction and summary extraction based on the target manual content to obtain the analysis manual content; obtain multiple long text fragments based on the analysis manual content, input the long text fragments into the target document large model for segmentation processing, and obtain the maintenance manual content based on the processing results.

[0051] Optionally, the answer recall module is also used to determine the manual semantic information based on the analyzed manual content; divide the analyzed manual content into multiple long text segments according to the manual semantic information; input each long text segment into the target document large model for segmentation processing to obtain a processing result; and obtain the maintenance manual content based on the processing result.

[0052] Optionally, the answer recall module is also used to convert the historical fault data into a question-answer pair format to obtain question-answer pair data; generate a historical fault file based on the question-answer pair data; and generate a historical fault knowledge base based on the historical fault file.

[0053] Optionally, the answer recall module is also used to obtain maintenance keywords based on the content of the maintenance manual; generate a keyword table based on the maintenance keywords; import the keyword table and the historical fault knowledge base into an initial knowledge graph to obtain a fault maintenance knowledge graph.

[0054] Optionally, the intent recognition module is also used to import the query question into the target language model to obtain user intent when receiving a query question about the cause of a target device failure; determine whether to call a vocabulary resource library based on the user intent, and obtain an intent recognition result based on the determination result.

[0055] In addition, to achieve the above objectives, the present application also proposes a fault repair device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the fault repair method described above.

[0056] In addition, to achieve the above objectives, the present application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the fault repair method described above are implemented.

[0057] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, wherein the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the fault repair method described above are implemented.

[0058] One or more technical solutions proposed in this application have at least the following technical effects:

[0059] When receiving a query question about the cause of a target device failure, the solution of the present application performs intent recognition on the query question through the target language large model to obtain an intent recognition result; calls the vocabulary resource library according to the intent recognition result, and queries the fault repair knowledge graph generated based on the target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers; performs network resource search according to the recall answer to obtain a network search result of the fault classification; determines the fault classification according to the recall answer and the network search result, and generates a fault resolution recommendation according to the fault classification. In this way, it is possible to perform intent recognition according to the user's query question, and then retrieve answers based on the pre-established knowledge graph. At the same time, network search is combined to perform fault classification and generate the final solution, providing customized and accurate troubleshooting suggestions, while relying on the knowledge graph and large model to improve the efficiency and accuracy of fault repair. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0061] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0062] Figure 1 A flowchart of the first embodiment of the fault repair method of the present application is provided;

[0063] Figure 2 A schematic diagram of model evaluation optimization in an embodiment of a fault repair method of the present application;

[0064] Figure 3 A flowchart of the second embodiment of the fault repair method of the present application is provided;

[0065] Figure 4 This is a brief flowchart of an embodiment of a fault repair method of the present application;

[0066] Figure 5 This is a schematic diagram of the module structure of the fault repair device according to an embodiment of the present application;

[0067] Figure 6 Schematic diagram of the equipment structure of the hardware operating environment involved in the fault repair method in the embodiment of the present application.

[0068] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0069] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0070] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0071] The main solution of the embodiment of the present application is: when a query question about the cause of a target device failure is received, the query question is identified with intent through the target language big model to obtain an intent recognition result; the vocabulary resource library is called according to the intent recognition result, and the fault repair knowledge graph generated based on the target document big model is queried according to the target vocabulary obtained by the call to obtain multiple recall answers; network resource search is performed according to the recall answers to obtain network search results of fault classification; the fault classification is determined according to the recall answer and the network search results, and a fault resolution recommendation is generated according to the fault classification.

[0072] In this embodiment, for the convenience of description, the following description is made by taking the identification computer as the execution subject.

[0073] Due to the existing technology, the fault maintenance of various high-end mechanical equipment, vehicles, and technological products is a highly professional and complex task, which requires not only maintenance personnel to have solid theoretical knowledge and practical experience, but also to be able to quickly and accurately locate the cause of the fault and take effective measures. The traditional maintenance model relies on manual review of a large amount of technical information and past cases, which is a time-consuming and error-prone process.

[0074] The present application provides a solution, through which intent recognition is achieved based on user query questions, and then answers are retrieved based on a pre-established knowledge graph. At the same time, network search is combined for fault classification and final solution generation, providing customized and accurate troubleshooting suggestions, while relying on knowledge graphs and large models to improve the efficiency and accuracy of fault repair.

[0075] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a server, etc. The following takes a fault repair device or a computer with a corresponding program installed as an example to illustrate this embodiment and the following embodiments.

[0076] Based on this, the present application embodiment provides a fault repair method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the fault repair method of the present application.

[0077] In this embodiment, the fault repair method includes steps S10 to S40:

[0078] Step S10, when receiving a query question about the cause of a target device failure, performing intent recognition on the query question through a target language large model to obtain an intent recognition result;

[0079] It should be noted that first, the query question input by the user is received through the UI interface, then the user's intention analysis is performed based on the query question, and then it is determined whether to call the vocabulary resource library based on the result of the intention analysis.

[0080] In a feasible implementation, in order to accurately analyze user intent, step S10 includes: when receiving a query question about the cause of a target device failure, importing the query question into a target language model to obtain user intent; determining whether to call a vocabulary resource library based on the user intent, and obtaining an intent recognition result based on the determination result.

[0081] It should be understood that in response to the questions raised by the user, the system performs the intent recognition process, which is driven by a large language model to evaluate whether it is necessary to call a specialized vocabulary resource library. This judgment mechanism ensures a deep understanding of the user's query intent and lays the foundation for subsequent processing. Among them, the large language model is the target language large model, which is a pre-trained large model of any type.

[0082] In a feasible implementation, in order to accurately obtain the user's intention, when a query question about the cause of the target device failure is received, the query question is imported into the target language model to obtain the user's intention. The step includes: when a query question about the cause of the target device failure is received, a query prompt word is determined according to the query question; subject name information is extracted from the query prompt word; and the subject name information and the query question are imported into the target language model to obtain the user's intention.

[0083] In the specific implementation, keywords are extracted from user questions through the method of prompt words + large models, and the main names such as fault codes and part names are extracted. Subsequently, knowledge graph retrieval and full-text retrieval of data sources are performed based on these keywords.

[0084] Step S20, calling a vocabulary resource library according to the intention recognition result, and querying a fault maintenance knowledge graph generated based on a target document large model according to the target vocabulary obtained by the call, to obtain multiple recall answers;

[0085] It should be noted that after obtaining the intent recognition result, the vocabulary resource library is called according to the intent recognition result. The vocabulary resource library is also a pre-set and stored resource library, which stores all the tendency words corresponding to the user's intent, such as: repair, maintenance, inspection and other words.

[0086] It should be understood that after obtaining the target vocabulary, the target vocabulary is imported into the fault maintenance knowledge graph, so that multiple recall answers can be obtained.

[0087] In the specific implementation, vector similarity does not have the transitive feature. For example, in the vector space, if historical question A is similar to historical question B, and historical question B is similar to historical question C, it does not mean that historical question A is similar to historical question C. In contrast, the knowledge graph stores a large number of clear facts in the form of precise, updateable and interpretable knowledge triples (i.e. (entity)->[relationship]->(entity)). In order to ensure that relevant historical issues can be correctly retrieved, this paper proposes to increase the exploration of similar historical issues. In addition, the association of synonyms and synonyms of proper nouns is also implemented to improve the accuracy and comprehensiveness of retrieval.

[0088] In a feasible implementation, step S20 includes: calling a vocabulary resource library according to the intention recognition result to obtain multiple target words corresponding to the query question; querying a fault repair knowledge graph according to the target words to obtain multiple vocabulary-related questions; and obtaining multiple recall answers according to the vocabulary-related questions.

[0089] It should be noted that, first, the vocabulary resource library is called according to the intention recognition results, and multiple target words corresponding to the semantics of the query question can be queried. Then, based on the target words, they are brought into the fault maintenance knowledge graph for query. Using the constructed fault maintenance knowledge graph, the system can generate multiple sets of related questions for information recall. This strategy not only enriches the query dimension, but also promotes the joint learning of enterprise-specific terms by the vector model and the large language model, significantly improving the accuracy and pertinence of the recall.

[0090] Step S30, searching network resources according to the recalled answer to obtain network search results of fault classification;

[0091] It should be understood that after obtaining the recall answer, it is necessary to perform query and retrieval in another dimension based on the recall answer to enrich and improve the query results.

[0092] In a specific implementation, network resource retrieval refers to querying and retrieving information in a pre-set network resource library based on the recalled answers, thereby obtaining network retrieval results.

[0093] In a feasible implementation, in order to accurately perform network retrieval, step S30 includes: filtering the recalled answers through a target proxy node to obtain a filtering result; determining the troubleshooting chapter number corresponding to a preset maintenance manual based on the filtering result; and retrieving the target network resource based on the troubleshooting chapter number to obtain a network retrieval result.

[0094] It should be noted that after the recall phase is completed, an intelligent agent node is introduced for advanced screening. The agent is responsible for analyzing the recalled information fragments to identify whether they contain numbers pointing to troubleshooting chapters. Once the relevant chapter number is confirmed, the system immediately activates a special tool to retrieve the corresponding network resource URL. This step ensures that the information obtained is directly related to the specific troubleshooting guide, enhancing practicality.

[0095] It should be understood that since a single vector search is difficult to fully retrieve complex content, some historical faults may not be recalled. Therefore, this paper will use a variety of retrieval methods, including vector retrieval, graph exploration, full-text retrieval, and question-answer retrieval, to ensure that all relevant information is fully recalled. By combining these methods, the coverage and accuracy of the retrieval system can be improved, ensuring that all relevant historical faults can be effectively identified and processed.

[0096] In specific implementation, when hybrid retrieval is used, a large number of historical fault records may be recalled, which will provide too much irrelevant content to the large model, causing interference and increasing computing power consumption. To solve this problem, this paper introduces the bge-reranker reranking model to rerank the recalled historical faults. This model can effectively screen out several historical faults with the highest similarity, thereby reducing knowledge interference, optimizing model performance and reducing computing power consumption.

[0097] Step S40, determining the fault classification according to the recalled answer and the network search result, and generating a fault solution recommendation according to the fault classification.

[0098] It should be understood that the model further determines the specific category to which the fault belongs, combining the original question with the answer extracted from the knowledge base. This comprehensive analysis process not only improves the speed of fault identification, but also improves the accuracy of classification, providing users with more accurate fault solution recommendations. Through this series of strategy integration, the intelligence level of the fault diagnosis system and user experience are effectively improved.

[0099] It should be noted that the user interface of the solution in this embodiment is designed to be user-friendly, allowing users to accurately locate the required model and detailed DE (defect code) number through an intuitive selection box. This mechanism ensures that the system can provide customized and accurate troubleshooting suggestions for each specific model, avoiding generalities and getting to the heart of the problem.

[0100] It should be understood that the system can also make full use of the advantages of big data, dynamically summarize and analyze user consultation logs, and instantly extract frequently occurring hot issues and display them in a clear manner. Users only need to click on these hot issues to get verified answers immediately, which greatly shortens the problem-solving cycle and achieves a significant leap in efficiency.

[0101] In specific implementation, the system can also record the consultation history of each user, which is not only convenient for individual follow-up inquiries and tracking, but also opens up new ways for team collaboration and knowledge inheritance. Users can easily share their consultation records with other maintenance engineers or managers, which promotes the efficient flow of information and experience sharing among teams, and jointly promotes the continuous improvement of maintenance efficiency. In summary, the auxiliary system based on large models has brought unprecedented efficiency improvements to the maintenance field through multiple optimizations of intelligence, personalization and collaboration.

[0102] It should be noted that if Figure 2As shown, in order to optimize the model and the entire process, after step S40, the data set can be tested four times using the RAGAS evaluation framework to comprehensively score from multiple dimensions, including context relevance, context recall, answer accuracy, and answer relevance. The final results show that the overall accuracy has been significantly improved from the original 60% to about 90%.

[0103] This embodiment provides a fault repair method. When receiving a query question about the cause of a target device fault, the query question is identified by the target language big model to obtain an intent recognition result; the vocabulary resource library is called according to the intent recognition result, and the fault repair knowledge graph generated based on the target document big model is queried according to the target vocabulary obtained by the call to obtain multiple recall answers; network resources are searched according to the recall answers to obtain network search results for fault classification; the fault classification is determined according to the recall answers and the network search results, and a fault solution recommendation is generated according to the fault classification. In this way, intent recognition is achieved according to the user's query question, and then the answer is retrieved based on the pre-established knowledge graph. At the same time, network search is combined to perform fault classification and generate the final solution, providing customized and accurate troubleshooting suggestions, and relying on the knowledge graph and big model to improve the efficiency and accuracy of fault repair.

[0104] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 3 Before step S20, the fault repair method further includes steps S201 to S203:

[0105] Step S201, dividing the content of a preset maintenance manual into a plurality of long text segments, and inputting the long text segments into a target document macro model for segmentation processing, and obtaining the maintenance manual content according to the processing results;

[0106] It should be noted that, firstly, the text content of the preset maintenance manual is divided into text segments to obtain multiple long text segments, and then the long text segments are imported into the target document big model, segmented by the target document big model, and then summarized and information extracted to obtain the maintenance manual content.

[0107] In a feasible implementation, in order to extract long text fragments, step S201 includes: obtaining a text file of a preset maintenance manual, and performing content extraction based on the text file to obtain the target manual content, wherein the content extraction includes paragraph, picture, table and page number extraction; performing label extraction and summary extraction based on the target manual content to obtain analysis manual content; and obtaining multiple long text fragments based on the analysis manual content.

[0108] It should be understood that the system uses the layout recognition technology to conduct an in-depth analysis of the Boeing manual content provided. First, the content such as paragraphs, pictures, tables and page numbers in the PDF file are extracted in a semantic understanding manner.

[0109] In the specific implementation, after content extraction, label extraction and summary extraction are performed on the paragraph content to obtain the analysis manual content, thereby improving the readability and retrieval efficiency of the document content. Finally, the text is divided according to the analysis manual content to obtain multiple long text fragments.

[0110] In a feasible implementation, in order to perform segmentation processing, the step of dividing the content of the preset maintenance manual into multiple long text segments includes: determining the manual semantic information based on the analyzed manual content; dividing the analyzed manual content into multiple long text segments based on the manual semantic information.

[0111] It should be noted that the analysis manual content is divided into multiple long text segments to ensure the semantic integrity of each segment. These segments are then input into the large model for intelligent segmentation processing to ensure data integrity and consistency. Finally, the document is formed according to the processing results to achieve accurate information extraction and efficient document management, thereby improving the efficiency of the maintenance manual. Among them, the target document large model here is a 32K document large model, or other large models that can achieve this function, which is not limited in this embodiment.

[0112] Step S202, generating a historical fault knowledge base according to the historical fault data;

[0113] It should be understood that the historical fault data is first extracted, which includes the historical faults and maintenance information records and data of the current equipment, and may also include the user's query records.

[0114] In a feasible implementation, in order to generate a more accurate and easy-to-query knowledge base, step S202 includes: converting the historical fault data into a question-answer pair format to obtain question-answer pair data; generating a historical fault file based on the question-answer pair data; and generating a historical fault knowledge base based on the historical fault file.

[0115] In the specific implementation, the system can automatically process historical fault measures data through built-in tools. First, data collation is performed: the historical fault data is uniformly collated and converted into the form of question-answer pairs. Specifically, the historical fault data is converted into a combination of paired questions and answers, and each combination of questions and answers is a question-answer pair. Then file generation is performed: the collated data generates a new historical fault file for subsequent management and query. Finally, the knowledge base is entered: the generated historical fault file is entered into the historical fault knowledge base as a data source for historical fault measures, providing a reference for future fault diagnosis.

[0116] Step S203, generating a fault maintenance knowledge graph based on the historical fault knowledge base and the maintenance manual content.

[0117] It should be noted that after obtaining the historical fault knowledge base and maintenance manual content, the knowledge graph for fault maintenance is constructed. Specifically, keywords are first extracted, and then the initial knowledge graph is imported to form a knowledge graph for fault maintenance of the target equipment.

[0118] In a feasible implementation, in order to accurately construct the knowledge graph, step S203 includes: obtaining maintenance keywords based on the content of the maintenance manual; generating a keyword table based on the maintenance keywords; importing the keyword table and the historical fault knowledge base into the initial knowledge graph to obtain a fault maintenance knowledge graph.

[0119] It should be understood that by importing maintenance keywords into the knowledge graph through a table, the query tool function is enhanced. First, enter the keywords. According to the content of the maintenance manual, multiple maintenance keywords can be obtained. Then, the maintenance keyword table is imported into the knowledge graph to facilitate the system to call when querying. Then, historical fault data is entered, historical fault problems are entered into the knowledge graph, and the association between historical problems is established.

[0120] In the specific implementation, the fault maintenance knowledge graph also includes an intelligent query tool: the knowledge graph serves as a backup for the intelligent query tool, supporting the intelligent agent to call relevant data during the fault diagnosis and problem solving process.

[0121] It should be noted that, taking aircraft maintenance as an example, Figure 4 The figure shows a flow chart, which can realize answer retrieval with the help of knowledge graph and knowledge base of historical fault data.

[0122] This embodiment divides the content of the preset maintenance manual into multiple long text segments, and inputs the long text segments into the target document large model for segmentation processing, and obtains the maintenance manual content according to the processing results; generates a historical fault knowledge base according to historical fault data; and generates a fault maintenance knowledge graph according to the historical fault knowledge base and the maintenance manual content. In this way, the construction of the knowledge graph is achieved by combining historical data and maintenance manuals, so that subsequent question queries can quickly query the closest and most accurate answers.

[0123] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the fault repair method of the present application. More simple transformations based on this technical concept are all within the protection scope of the present application.

[0124] This application also provides a fault repair device, please refer to Figure 5 , the fault repair device comprises:

[0125] The intention recognition module 10 is used to perform intention recognition on the query question about the cause of the target device failure through the target language large model when receiving the query question to obtain the intention recognition result.

[0126] The answer recall module 20 is used to call the vocabulary resource library according to the intention recognition result, and query the fault maintenance knowledge graph generated based on the target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers.

[0127] The network search module 30 is used to search network resources according to the recalled answers to obtain network search results of fault classification.

[0128] The solution generation module 40 is used to determine the fault classification according to the recalled answer and the network search result, and generate a fault solution recommendation according to the fault classification.

[0129] In one embodiment, the answer recall module 20 is also used to divide the content of a preset maintenance manual into multiple long text segments, and input the long text segments into the target document model for segmentation processing, and obtain the maintenance manual content according to the processing results; generate a historical fault knowledge base according to historical fault data; and generate a fault maintenance knowledge graph according to the historical fault knowledge base and the maintenance manual content.

[0130] In one embodiment, the answer recall module 20 is also used for the step of dividing the content of the preset maintenance manual into multiple long text segments, including: obtaining a text file of the preset maintenance manual, and performing content extraction based on the text file to obtain the target manual content, the content extraction including paragraph, picture, table and page number extraction; performing label extraction and summary extraction based on the target manual content to obtain analysis manual content; and obtaining multiple long text segments based on the analysis manual content.

[0131] In one embodiment, the answer recall module 20 is further used to determine manual semantic information based on the analyzed manual content; and divide the analyzed manual content into multiple long text segments based on the manual semantic information.

[0132] In one embodiment, the answer recall module 20 is further used to convert the historical fault data into a question-answer pair format to obtain question-answer pair data; generate a historical fault file based on the question-answer pair data; and generate a historical fault knowledge base based on the historical fault file.

[0133] In one embodiment, the answer recall module 20 is further used to obtain maintenance keywords based on the content of the maintenance manual; generate a keyword table based on the maintenance keywords; import the keyword table and the historical fault knowledge base into the initial knowledge graph to obtain a fault maintenance knowledge graph.

[0134] In one embodiment, the intention recognition module 10 is also used to import the query question into the target language model to obtain the user intention when receiving the query question about the cause of the target device failure; determine whether to call the vocabulary resource library according to the user intention, and obtain the intention recognition result according to the determination result.

[0135] In one embodiment, the intention recognition module 10 is also used to determine the query prompt word according to the query question when receiving the query question about the cause of the target device failure; extract the subject name information in the query prompt word; import the subject name information and the query question into the target language model to obtain the user intention.

[0136] In one embodiment, the answer recall module 20 is further used to call a vocabulary resource library according to the intention recognition result to obtain multiple target words corresponding to the query question; query the fault repair knowledge graph according to the target words to obtain multiple vocabulary-related questions; and obtain multiple recall answers according to the vocabulary-related questions.

[0137] In one embodiment, the network retrieval module 30 is also used to filter the recalled answers through a target proxy node to obtain a filtering result; determine the troubleshooting chapter number corresponding to a preset maintenance manual based on the filtering result; and retrieve the target network resource based on the troubleshooting chapter number to obtain a network retrieval result.

[0138] When receiving a query question about the cause of a target device failure, this embodiment uses the target language big model to perform intent recognition on the query question to obtain an intent recognition result; calls the vocabulary resource library based on the intent recognition result, and queries the fault repair knowledge graph generated based on the target document big model based on the target vocabulary obtained by the call to obtain multiple recall answers; performs a network resource search based on the recall answer to obtain a network search result of the fault classification; determines the fault classification based on the recall answer and the network search result, and generates a fault solution recommendation based on the fault classification. In this way, intent recognition is achieved based on the user's query question, and then answers are retrieved based on the pre-established knowledge graph. At the same time, network search is combined to perform fault classification and generate the final solution, providing customized and accurate troubleshooting suggestions, while relying on the knowledge graph and big model to improve the efficiency and accuracy of fault repair.

[0139] The fault repair device provided by the present application adopts the fault repair method in the above embodiment, which can solve the technical problem that equipment fault troubleshooting is time-consuming and error-prone. Compared with the prior art, the beneficial effects of the fault repair device provided by the present application are the same as the beneficial effects of the fault repair method provided by the above embodiment, and other technical features in the fault repair device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0140] The present application provides a fault repair device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the fault repair method in the above-mentioned embodiment 1.

[0141] Reference below Figure 6, which shows a schematic diagram of the structure of a fault repair device suitable for implementing the embodiment of the present application. The fault repair device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The fault repair device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0142] like Figure 6 As shown, the fault repair device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the fault repair device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the fault repair device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows a fault repair device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0143] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0144] The fault repair device provided by the present application adopts the fault repair method in the above embodiment, which can solve the technical problem that equipment fault troubleshooting is time-consuming and error-prone. Compared with the prior art, the beneficial effects of the fault repair device provided by the present application are the same as the beneficial effects of the fault repair method provided by the above embodiment, and other technical features in the fault repair device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0145] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0146] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0147] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the fault repair method in the above-mentioned embodiment.

[0148] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0149] The computer-readable storage medium may be included in the fault repair device; or may exist independently without being assembled into the fault repair device.

[0150] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the fault repair device, the fault repair device: when receiving a query question about the cause of the target device fault, performs intent recognition on the query question through the target language big model to obtain an intent recognition result; calls a vocabulary resource library according to the intent recognition result, and queries the fault repair knowledge graph generated based on the target document big model according to the target vocabulary obtained by the call to obtain multiple recall answers; performs network resource search according to the recall answers to obtain network search results of fault classification; determines the fault classification according to the recall answer and the network search results, and generates a fault solution recommendation according to the fault classification.

[0151] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0152] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0153] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0154] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned fault repair method, and can solve the technical problem that equipment fault troubleshooting is time-consuming and error-prone. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the fault repair method provided by the above-mentioned embodiment, and will not be repeated here.

[0155] The present application also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the steps of the fault repair method described above are implemented.

[0156] The computer program product provided by the present application can solve the technical problem that equipment fault troubleshooting is time-consuming and error-prone. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the fault repair method provided by the above embodiment, which will not be repeated here.

[0157] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

[0158] The present invention discloses A1. a fault repair method, the method comprising:

[0159] When receiving a query question about the cause of a target device failure, performing intent recognition on the query question through the target language large model to obtain an intent recognition result;

[0160] Calling a vocabulary resource library according to the intention recognition result, and querying a fault maintenance knowledge graph generated based on a target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers;

[0161] Perform network resource search according to the recalled answer to obtain network search results of fault classification;

[0162] A fault classification is determined according to the recall answer and the network search result, and a fault solution recommendation is generated according to the fault classification.

[0163] A2. The method as described in A1, before the step of calling the vocabulary resource library according to the intention recognition result and querying the fault repair knowledge graph generated based on the target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers, also includes:

[0164] Divide the content of the preset maintenance manual into multiple long text segments, input the long text segments into the target document macro model for segmentation processing, and obtain the maintenance manual content according to the processing results;

[0165] Generate a historical fault knowledge base based on historical fault data;

[0166] A fault maintenance knowledge graph is generated based on the historical fault knowledge base and the maintenance manual content.

[0167] A3. As described in the method of A2, the step of dividing the content of the preset maintenance manual into a plurality of long text segments comprises:

[0168] Obtaining a text file of a preset maintenance manual, and extracting the content of the target manual according to the text file, wherein the content extraction includes paragraph, picture, table and page number extraction;

[0169] Perform label extraction and summary extraction according to the target manual content to obtain the analysis manual content;

[0170] A plurality of long text segments are obtained based on the analysis of the manual content.

[0171] A4. As described in the method of A3, the step of obtaining a plurality of long text segments according to the content of the analysis manual comprises:

[0172] Determining manual semantic information based on analyzing the manual content;

[0173] The analyzed manual content is divided into a plurality of long text segments according to the manual semantic information.

[0174] A5. As described in the method of A2, the step of generating a historical fault knowledge base based on historical fault data comprises:

[0175] Converting the historical fault data into a question-answer pair format to obtain question-answer pair data;

[0176] Generate a historical fault file based on the question and answer data;

[0177] A historical fault knowledge base is generated according to the historical fault files.

[0178] A6. As described in the method of A2, the step of generating a fault maintenance knowledge graph based on the historical fault knowledge base and the maintenance manual content includes:

[0179] Obtain maintenance keywords according to the maintenance manual content;

[0180] generating a keyword table according to the maintenance keywords;

[0181] The keyword table and the historical fault knowledge base are imported into the initial knowledge graph to obtain a fault maintenance knowledge graph.

[0182] A7. As described in the method of A1, when receiving a query question about the cause of the target device failure, the step of performing intent recognition on the query question through the target language large model to obtain the intent recognition result includes:

[0183] When receiving a query question about the cause of a target device failure, importing the query question into the target language macro model to obtain the user intention;

[0184] It is determined whether to call a vocabulary resource library according to the user intention, and an intention recognition result is obtained according to the determination result.

[0185] A8. As described in the method of A7, when receiving a query question about the cause of the target device failure, the step of importing the query question into the target language large model to obtain the user's intention includes:

[0186] When receiving a query question about the cause of a target device failure, determining a query prompt word according to the query question;

[0187] Extracting subject name information from the query prompt word;

[0188] The subject name information and the query question are imported into the target language macro model to obtain the user intention.

[0189] A9. As described in the method of A1, the step of calling the vocabulary resource library according to the intention recognition result, and querying the fault repair knowledge graph generated based on the target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers includes:

[0190] Calling a vocabulary resource library according to the intention recognition result to obtain a plurality of target words corresponding to the query question;

[0191] According to the target vocabulary, a fault maintenance knowledge graph is searched to obtain a plurality of vocabulary-related questions;

[0192] A plurality of recalled answers are obtained according to the vocabulary-related questions.

[0193] A10. As described in the method of A1, the step of performing a network resource search based on the recalled answer to obtain a network search result of the fault classification includes:

[0194] Screening the recalled answers through the target proxy node to obtain a screening result;

[0195] Determine the troubleshooting chapter number corresponding to the preset maintenance manual according to the screening result;

[0196] The target network resource is searched according to the troubleshooting chapter number to obtain a network search result.

[0197] The present invention also discloses B11. a fault repair device, the device comprising:

[0198] An intent recognition module is used to, when receiving a query question about the cause of a target device failure, perform intent recognition on the query question through a target language large model to obtain an intent recognition result;

[0199] An answer recall module is used to call a vocabulary resource library according to the intention recognition result, and query the fault maintenance knowledge graph generated based on the target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers;

[0200] A network search module, used to search network resources according to the recalled answers to obtain network search results for fault classification;

[0201] A solution generation module is used to determine the fault classification according to the recalled answer and the network search result, and to generate a recommended fault solution according to the fault classification.

[0202] B12. In the device as described in B11, the answer recall module is also used to divide the content of a preset maintenance manual into multiple long text segments, and input the long text segments into the target document large model for segmentation processing, and obtain the content of the maintenance manual according to the processing results; generate a historical fault knowledge base according to historical fault data; and generate a fault maintenance knowledge graph according to the historical fault knowledge base and the maintenance manual content.

[0203] B13. In the device as described in B12, the answer recall module is also used to obtain a text file of a preset maintenance manual, and perform content extraction based on the text file to obtain the target manual content, the content extraction includes paragraph, picture, table and page number extraction; perform label extraction and summary extraction based on the target manual content to obtain the analysis manual content; obtain multiple long text fragments based on the analysis manual content, input the long text fragments into the target document large model for segmentation processing, and obtain the maintenance manual content based on the processing results.

[0204] B14. In the device as described in B13, the answer recall module is also used to determine the manual semantic information based on the analyzed manual content; divide the analyzed manual content into multiple long text segments based on the manual semantic information; input each long text segment into the target document large model for segmentation processing to obtain a processing result; and obtain the maintenance manual content based on the processing result.

[0205] B15. In the device as described in B12, the answer recall module is also used to convert the historical fault data into a question-answer pair format to obtain question-answer pair data; generate a historical fault file based on the question-answer pair data; and generate a historical fault knowledge base based on the historical fault file.

[0206] B16. In the device as described in B12, the answer recall module is also used to obtain maintenance keywords based on the content of the maintenance manual; generate a keyword table based on the maintenance keywords; import the keyword table and the historical fault knowledge base into the initial knowledge graph to obtain the fault maintenance knowledge graph.

[0207] B17. In the device as described in B11, the intention recognition module is also used to import the query question about the cause of the target device failure into the target language model to obtain the user intention; determine whether to call the vocabulary resource library based on the user intention, and obtain the intention recognition result based on the determination result.

[0208] The present invention also discloses C18. A fault repair device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the fault repair method as described above.

[0209] The present invention also discloses D19. a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the fault repair method as described above are implemented.

[0210] The present invention also discloses E20. a computer program product, the computer program product comprising a computer program, and the computer program implements the steps of the fault repair method described above when executed by a processor.

Claims

1. A fault repair method, characterized in that: The method includes: When receiving a query question about the cause of a target device failure, performing intent recognition on the query question through the target language big model to obtain an intent recognition result; Calling a vocabulary resource library according to the intention recognition result, and querying a fault maintenance knowledge graph generated based on a target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers; Perform network resource search according to the recalled answer to obtain network search results of fault classification; A fault classification is determined according to the recalled answers and the network search results, and a fault resolution recommendation is generated according to the fault classification.

2. The method according to claim 1, characterized in that Before the step of calling the vocabulary resource library according to the intention recognition result and querying the fault maintenance knowledge graph generated based on the target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers, the step further includes: Divide the content of the preset maintenance manual into multiple long text segments, input the long text segments into the target document macro model for segmentation processing, and obtain the maintenance manual content according to the processing results; Generate a historical fault knowledge base based on historical fault data; A fault maintenance knowledge graph is generated based on the historical fault knowledge base and the maintenance manual content.

3. The method according to claim 2, characterized in that The step of dividing the content of the preset maintenance manual into a plurality of long text segments comprises: Obtaining a text file of a preset maintenance manual, and extracting the content of the target manual according to the text file, wherein the content extraction includes paragraph, picture, table and page number extraction; Perform label extraction and summary extraction according to the target manual content to obtain the analysis manual content; A plurality of long text segments are obtained based on the analysis of the manual content.

4. The method according to claim 3, characterized in that The step of obtaining a plurality of long text segments according to the content of the analysis manual comprises: Determining manual semantic information based on analyzing the manual content; The analyzed manual content is divided into a plurality of long text segments according to the manual semantic information.

5. The method according to claim 2, characterized in that The step of generating a historical fault knowledge base according to historical fault data comprises: Converting the historical fault data into a question-answer pair format to obtain question-answer pair data; Generate a historical fault file based on the question and answer data; A historical fault knowledge base is generated according to the historical fault files.

6. The method according to claim 2, characterized in that The step of generating a fault maintenance knowledge graph according to the historical fault knowledge base and the maintenance manual content comprises: Obtain maintenance keywords according to the maintenance manual content; generating a keyword table according to the maintenance keywords; The keyword table and the historical fault knowledge base are imported into the initial knowledge graph to obtain a fault maintenance knowledge graph.

7. A fault repair device, characterized in that: The device comprises: An intention recognition module is used to, when receiving a query question about the cause of a target device failure, perform intention recognition on the query question through a target language large model to obtain an intention recognition result; An answer recall module is used to call a vocabulary resource library according to the intention recognition result, and query the fault maintenance knowledge graph generated based on the target document large model according to the target vocabulary obtained by the call to obtain multiple recall answers; A network search module, used to search network resources according to the recalled answers to obtain network search results for fault classification; A solution generation module is used to determine the fault classification according to the recalled answer and the network search result, and to generate a recommended fault solution according to the fault classification.

8. A fault repair device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the fault repair method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the fault repair method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the fault repair method according to any one of claims 1 to 6 are implemented.

Images