Aviation maintenance knowledge graph construction method, system, equipment and medium

Abstract

The invention discloses an aviation maintenance knowledge graph construction method, system and device and a medium, and the method comprises the steps: firstly generating an initial knowledge graph according to an aviation maintenance text through a large language model, and receiving a newly input aviation maintenance text data stream through a message queue for incremental knowledge extraction; updating the initial knowledge graph according to the incrementally extracted knowledge, then acquiring user feedback data to optimize the updated knowledge graph, generating an optimized knowledge graph, and extracting structured knowledge from image and / or video data in the aviation maintenance field by adopting a multi-modal large language model to obtain an optimized knowledge graph; and fusing the extracted knowledge with the optimized knowledge graph, and finally providing an intelligent question and answer service for the user based on the fused knowledge graph. Wherein the user feedback data generated in the intelligent question and answer service process is used for performing next optimization on the fused knowledge graph. According to the invention, the construction efficiency, timeliness and integrity of the aviation maintenance knowledge graph can be improved.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to a method, system, device and medium for constructing an aviation maintenance knowledge graph. Background Technology

[0002] Aviation maintenance support is a core component ensuring flight safety and operational efficiency. It involves aircraft maintenance manuals, fault isolation manuals, service bulletins, airworthiness directives, and a vast amount of maintenance records, drawings, and video data, forming a multimodal, massive, and rapidly iterating knowledge system. Knowledge graphs, as semantic networks capable of effectively expressing complex relationships, have been attempted for application in aviation maintenance to improve the structuring and usability of knowledge. However, current knowledge graph construction technologies in aviation maintenance still struggle to meet the demands of large-scale, dynamic, and high-precision maintenance knowledge management. Existing construction methods mainly focus on the following two categories, both of which have significant limitations: Firstly, there is the knowledge extraction method based on human expert experience. This method relies on experts in the aviation maintenance field to manually read and sort through massive amounts of technical documents, manually identify and extract knowledge elements, and then input them one by one into the knowledge graph. Although this method can ensure a certain degree of accuracy based on expert experience, manual processing is time-consuming and labor-intensive, and the graphing efficiency is extremely low, making it difficult to support the rapid construction of large-scale knowledge graphs. Moreover, the knowledge extraction results are easily affected by the subjective cognition and experience differences of experts, and there is a risk of knowledge omissions or inconsistent labeling, which further limits the completeness and reliability of the knowledge graph.

[0003] Secondly, there are information extraction techniques based on rules or traditional machine learning. This approach mainly achieves automatic knowledge extraction by pre-setting domain rules (such as regular expressions) or training supervised learning models (such as Conditional Random Fields (CRF) or fine-tuning pre-trained language models like BERT). However, rule formulation requires an exhaustive review of domain knowledge, and training supervised learning models requires large-scale, high-quality labeled datasets. In the aviation maintenance field, new aircraft models and fault modes are constantly emerging, making the acquisition of suitable labeled data extremely costly. Furthermore, rules are prone to failure and models are prone to misjudgment when faced with unseen entity types (such as new parts), relational patterns, or variations in terminology, making it difficult to handle the complex semantics in maintenance texts. Simultaneously, it cannot solve problems such as contextual understanding and semantic disambiguation, easily leading to knowledge redundancy or broken connections in the knowledge graph. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, equipment, and medium for constructing an aviation maintenance knowledge graph, thereby improving the efficiency, timeliness, and completeness of the construction of the aviation maintenance knowledge graph.

[0005] To achieve the above objectives, this invention provides a method for constructing an aviation maintenance knowledge graph, comprising: A large language model is used to extract structured knowledge from aviation maintenance texts. The extracted knowledge is then fused with the domain ontology to generate an initial knowledge graph. The system receives new input aviation maintenance text data streams via a message queue, performs incremental knowledge extraction, updates the initial knowledge graph based on the incrementally extracted knowledge, and generates an updated knowledge graph. Collect user feedback data to optimize the updated knowledge graph and generate an optimized knowledge graph; A multimodal large language model is used to extract structured knowledge from image and / or video data in the field of aviation maintenance, and the extracted knowledge is fused with an optimized knowledge graph. Based on the fused knowledge graph, intelligent question-answering services are provided to users; the user feedback data generated during the intelligent question-answering service is used to further optimize the fused knowledge graph.

[0006] Preferably, the step of extracting structured knowledge from historical aviation maintenance texts using a large language model, and then fusing the extracted knowledge with a domain ontology to generate an initial knowledge graph, includes: The aircraft maintenance text is segmented into text units; Constructing prompt words in the aviation domain guides a large language model to extract structured knowledge from the text units in a few-shot or zero-shot manner; wherein, the structured knowledge includes named entities, relationships between entities, and events; The extracted knowledge is disambiguated by using the large language model to identify the same entity. The disambiguated knowledge is then integrated with the aviation maintenance domain ontology to generate the initial knowledge graph.

[0007] Preferably, the structure receives newly input aviation maintenance text data streams through a message queue, performs incremental knowledge extraction, updates the initial knowledge graph based on the incrementally extracted knowledge, and generates an updated knowledge graph, including: Receive new input aircraft maintenance text data streams via message queue; The large language model is used to perform incremental knowledge extraction on the aviation maintenance text data stream to obtain structured knowledge fragments; Add a time attribute to the structured knowledge fragment; Structured knowledge fragments with added time attributes are fused into the initial knowledge graph through a graph database interface in an incremental update manner to generate an updated knowledge graph.

[0008] Preferably, the step of collecting user feedback data to optimize the updated knowledge graph and generate an optimized knowledge graph includes: Collect user feedback data, generate reverse prompts based on the user feedback data, and train the large language model to identify erroneous knowledge extraction patterns; When preset conditions are met, the large language model is triggered to review the target knowledge entries in the updated knowledge graph that are related to the user feedback data. Based on the review results, revision suggestions are generated for the target knowledge items; The updated knowledge graph is optimized based on the proposed corrections to generate an optimized knowledge graph.

[0009] Preferably, the step of extracting structured knowledge from image and / or video data in the field of aviation maintenance using a multimodal large language model, and fusing the extracted knowledge with an optimized knowledge graph, includes: When the input is image data from the field of aircraft maintenance. A multimodal large language model is used to identify the location of components, connections, or abnormal regions in an image, generate structured knowledge describing the components, relationships, or abnormalities, and link the generated structured knowledge with entities representing the components in an optimized knowledge graph. When the input is video data from the aviation maintenance field... By using a multimodal large language model to analyze video action sequences, the structured elements of maintenance operations are identified, and a structured maintenance event chain is generated. The event chain is then stored as a standard operating procedure in an optimized knowledge graph.

[0010] Preferably, the provision of intelligent question-answering services to users based on the fused knowledge graph includes: Receive questions from users in natural language. A large language model is used as the reasoning engine to perform semantic parsing on the question, and queries and causal inferences are performed based on the fused knowledge graph to generate answers, reasoning explanations, and the knowledge sources on which the answers are based.

[0011] Preferably, the time attribute includes the effective date, the expiration date, or revision information.

[0012] To achieve the above objectives, the present invention also provides an aviation maintenance knowledge graph construction system, comprising: The initial knowledge graph construction module is used to extract structured knowledge from aviation maintenance texts using a large language model, and then integrate the extracted knowledge with the domain ontology to generate the initial knowledge graph. The knowledge graph update module is used to receive newly input aviation maintenance text data streams through a message queue, perform incremental knowledge extraction, update the initial knowledge graph based on the incrementally extracted knowledge, and generate an updated knowledge graph. The knowledge graph optimization module is used to collect user feedback data to optimize the updated knowledge graph and generate an optimized knowledge graph. The knowledge graph fusion module is used to extract structured knowledge from image and / or video data in the field of aviation maintenance using a multimodal large language model, and then fuse the extracted knowledge with the optimized knowledge graph. The knowledge graph application module is used to provide intelligent question-answering services to users based on the fused knowledge graph; the user feedback data generated during the intelligent question-answering service is used to further optimize the fused knowledge graph.

[0013] To achieve the above objectives, the present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the aviation maintenance knowledge graph construction method as described above.

[0014] To achieve the above objectives, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the aviation maintenance knowledge graph construction method described in any of the above embodiments.

[0015] Compared with existing technologies, this invention provides a method, system, device, and medium for constructing an aviation maintenance knowledge graph. It extracts textual knowledge through a large language model and integrates it with a domain ontology to generate an initial graph, reducing manual intervention and addressing issues such as low efficiency, subjective bias, weak semantic disambiguation in traditional technologies, and reliance on large amounts of labeled data. It utilizes incremental updates via message queues to address the insufficient timeliness of static graphs; it incorporates closed-loop optimization based on user feedback to improve knowledge completeness and reliability; and it compensates for missing multimodal information by integrating image / video knowledge through a multimodal large model. This invention can improve the construction efficiency, timeliness, and completeness of aviation maintenance knowledge graphs, providing precise support for aircraft maintenance personnel and ensuring flight safety and operational efficiency. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a method for constructing an aviation maintenance knowledge graph according to an embodiment of the present invention; Figure 2This is a structural block diagram of an aviation maintenance knowledge graph construction system provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of a terminal device provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] See Figure 1 , Figure 1 This is a flowchart of a method for constructing an aviation maintenance knowledge graph according to an embodiment of the present invention. The method includes steps S1 to S5: Step S1: Use a large language model to extract structured knowledge from aviation maintenance texts, and integrate the extracted knowledge with the domain ontology to generate an initial knowledge graph. In an optional embodiment, step S1 includes steps S101 to S104: Step S101: Segment the aircraft maintenance text to form text units; Step S102: Construct prompt words in the aviation domain to guide the large language model to extract structured knowledge from the text units in a few-sample or zero-sample manner; wherein, the structured knowledge includes named entities, relationships between entities, and events; Step S103: Use the large language model to perform same entity recognition on the extracted knowledge in order to disambiguate the knowledge; Step S104: Integrate the disambiguated knowledge with the aviation maintenance domain ontology to generate the initial knowledge graph.

[0020] Understandably, unstructured aviation maintenance texts include documents such as Aircraft Maintenance Manuals (AMMs), Troubleshooting Manuals (TSMs), Parts Catalogs (IPCs), Flight Logs, Maintenance Work Orders (STOs), Service Notices (SBs), Airworthiness Directives (ADs), and Technical Logs. These unstructured aviation maintenance texts are used as input to a Large Language Model (LLM) for structured knowledge extraction.

[0021] Specifically, the original document is segmented into logical paragraphs or sentence units, and standardized prompts are designed to guide the large language model to perform specific tasks, such as: "Please identify all relevant aircraft components, fault codes, maintenance operations, tools used, and personnel qualification requirements from the following aviation maintenance text, and output the entities and their categories in JSON format." Or, “Please determine what specialized tools are required for the operation of ‘replacing the left IDG’ and indicate its relationship with ‘generator’.”

[0022] Then, by leveraging the contextual modeling capabilities of the large language model, multiple expressions of the same entity in different documents (such as abbreviations, full names, and model aliases) are identified, entity linking and disambiguation are achieved, and an ontology from the aviation maintenance field is introduced as a reference framework (such as the S1000D standard). The large language model determines whether newly extracted entities should be classified into existing categories or suggests adding new categories. At the same time, a preliminary comparative analysis is performed on conflicting information (such as inconsistent descriptions of the same part), and items are marked for manual review.

[0023] In this embodiment of the invention, the final large language model extraction result includes three basic elements: named entities, relationships between entities, and events.

[0024] For example, when the input text is "Maintenance inspection revealed a malfunction in the air circulator of the aircraft's air conditioning system, causing uncontrolled cabin temperature; the component has been replaced.", the extracted results include: 1. Named Entities: System: Air Conditioning System; Component: Air Circulator; Component: Cabin Temperature Control System; 2. Relationship between entities: The air circulator belongs to the air conditioning system; the air conditioning system controls the cabin temperature control system; 3. Event: Event type: Component failure and replacement failure; Component: Air circulator; System: Air conditioning system; Impact: Cabin temperature out of control; Remedial measures: Replacement.

[0025] Preferably, the accuracy of a large language model on a specific subtask can be improved by adding a small number of labeled examples to the prompts.

[0026] It's important to note that under conditions of few or even zero samples, the powerful generation and reasoning capabilities of large models can be fully utilized. Through techniques such as self-instruction, semantic expansion, and high-quality sample synthesis, domain-specific labeled datasets can be automatically constructed. For example, using a small number of real samples or domain knowledge as "seeds," an open-source large model (such as Qwen) can be driven to automatically generate high-quality synthetic data across multiple rounds and scenarios according to a preset extraction task format, covering typical contexts such as different equipment types, fault phenomena, and maintenance measures. Subsequently, the generated samples can undergo automatic or minimal manual review to form a high-confidence pseudo-label set, which can be used for further training or fine-tuning of the model. This achieves a transitional adaptation from zero samples to few samples, significantly reducing manual labeling costs while improving the model's extraction capabilities and generalization performance in the aviation maintenance domain. With sufficient labeled data, the open-source large model (Qwen) can be fine-tuned for domain adaptation, further improving its extraction accuracy in the aviation maintenance context. Fine-tuning targets include sub-tasks such as Named Entity Recognition (NER), Relation Classification (RC), and Event Extraction (EE).

[0027] Step S2: Receive newly input aviation maintenance text data streams through message queues, perform incremental knowledge extraction, update the initial knowledge graph based on the incrementally extracted knowledge, and generate an updated knowledge graph. In an optional embodiment, step S2 includes steps S201 to S204: Step S201: Receive the newly input aircraft maintenance text data stream through the message queue; Step S202: Use the large language model to perform incremental knowledge extraction on the aviation maintenance text data stream to obtain structured knowledge fragments; Step S203: Add a time attribute to the structured knowledge fragment; Step S204: The structured knowledge fragments with added time attributes are fused into the initial knowledge graph through the graph database interface in an incremental update manner to generate the updated knowledge graph.

[0028] For example, a message queue (such as Kafka) is built to receive newly released aviation maintenance document data streams (such as daily SB / AD pushes and real-time work order uploads). After preprocessing, the data is sent to a large language model engine to perform incremental knowledge extraction. The extraction results can be integrated into the existing knowledge graph through graph database APIs using "add, delete, and modify" operations, thus avoiding full reconstruction.

[0029] Preferably, the time attribute includes the effective date, the expiration date, or revision information.

[0030] For example, all knowledge nodes and edges in the knowledge graph are given a time attribute, which enables the large language model to parse time expressions in the text (such as "applicable from January 2025"), thereby automatically marking the validity period of knowledge. At the same time, it can support queries based on time slices, such as "query A320 flap inspection procedures valid before July 2024".

[0031] Step S3: Collect user feedback data to optimize the updated knowledge graph and generate an optimized knowledge graph; In an optional embodiment, step S3 includes steps S301 to S304: Step S301: Collect user feedback data, generate reverse prompts based on the user feedback data, and train the large language model to identify erroneous knowledge extraction patterns. Step S302: When the preset conditions are met, the large language model is triggered to review the target knowledge entries in the updated knowledge graph that are related to the user feedback data. Step S303: Based on the review results, generate correction suggestions for the target knowledge item; Step S304: Optimize the updated knowledge graph according to the correction suggestions to generate an optimized knowledge graph.

[0032] For example, feedback data (such as "inaccurate answer" or "missing steps") submitted by users through the interface when using knowledge graphs to query or obtain recommendations is collected; reverse prompts are generated based on the user feedback data, and a large language model is trained to recognize and avoid erroneous knowledge extraction patterns.

[0033] When preset conditions are met, the large language model is triggered to review or generate correction suggestions for target knowledge entries in the updated knowledge graph related to the user feedback data. For example, if an engineer queries "engine oil pressure abnormality troubleshooting steps" and reports "missing oil pressure sensor inspection steps," this type of feedback is collected and a reverse prompt is generated to train the model to identify similar omissions. Frequently reported knowledge entries are collected periodically (e.g., weekly) to trigger the review mechanism. During the review, the large language model is invoked to compare and evaluate the entry against technical data, generate correction suggestions, and prompt maintenance experts to review and confirm before updating the knowledge graph.

[0034] Step S4: Use a multimodal large language model to extract structured knowledge from image and / or video data in the field of aviation maintenance, and fuse the extracted knowledge with the optimized knowledge graph; In one alternative embodiment, step S4 includes: When the input is image data from the field of aircraft maintenance. A multimodal large language model is used to identify the location of components, connections, or abnormal regions in an image, generate structured knowledge describing the components, relationships, or abnormalities, and link the generated structured knowledge with entities representing the components in an optimized knowledge graph. When the input is video data from the aviation maintenance field... By using a multimodal large language model to analyze video action sequences, the structured elements of maintenance operations are identified, and a structured maintenance event chain is generated. The event chain is then stored as a standard operating procedure in an optimized knowledge graph.

[0035] It is worth noting that, in order to make full use of visual information in aircraft maintenance, this embodiment of the invention introduces a multimodal large language model (Multimodal LLM, MLLM) to break down the knowledge barriers between text and images / videos and to integrate text knowledge with image / video knowledge.

[0036] Specifically, the system inputs image files such as aircraft structural diagrams, circuit diagrams, and fault diagrams. MLLM (such as Qwen-VL or BLIP-2) is used to parse the image content, identify the locations of components, connections, and abnormal areas marked in the image, and then converts the image recognition results into structured triples and links them with entities in a text knowledge base. For example, when a photo of an engine combustion chamber crack is uploaded, the system can identify the crack location using MLLM and associate the identified crack location with the "combustion chamber component" entity in the knowledge graph. Finally, it returns relevant maintenance manual chapters, historical similar cases, and replacement procedures.

[0037] Specifically, by inputting maintenance operation instruction videos or on-site recordings, the system uses MLLM to analyze the action sequence frame by frame, identify key operation nodes (such as "removing the cover plate" and "measuring the gap"), involved objects (tools, parts), and sequential dependencies, and automatically generates structured event chains, which are stored in the knowledge graph as "standard operating procedures" nodes. This enables the knowledge graph to support bidirectional jumping between video clips and knowledge nodes, making it convenient for engineers to learn by comparison.

[0038] Step S5: Based on the fused knowledge graph, provide intelligent question answering services to users; wherein, the user feedback data generated during the intelligent question answering service process is used to further optimize the fused knowledge graph.

[0039] In one optional embodiment, step S5 includes: Receive questions from users in natural language. A large language model is used as the reasoning engine to perform semantic parsing on the question, and queries and causal inferences are performed based on the fused knowledge graph to generate answers, reasoning explanations, and the knowledge sources on which the answers are based.

[0040] For example, users can ask questions in natural language, such as "What are the possible causes of surge in the right engine of B787?" or "What tools and qualifications are needed to replace the main wheel?" At this time, a large language model can be called as an inference engine, which can be combined with the optimized knowledge graph to perform multi-hop queries, causal inference and explanation generation. The output not only includes the answer, but can also include source evidence, such as reference manual number, diagram link, historical case, etc. At the same time, user input can support voice input and output, thereby adapting to the complex environment of the maintenance site.

[0041] In summary, the aviation maintenance knowledge graph construction method provided by this invention employs a large language model to extract structured knowledge from aviation maintenance texts and integrates it with a domain ontology to generate an initial knowledge graph. This eliminates the need for domain experts to manually analyze documents, significantly reducing human intervention and solving the problems of time-consuming and labor-intensive manual extraction, as well as the susceptibility to subjective perception leading to knowledge omissions or inconsistent labeling. Furthermore, the semantic understanding capabilities of the large language model, combined with the domain ontology, effectively identify variations in terminology and eliminate contextual ambiguity, avoiding knowledge redundancy or fragmentation caused by insufficient semantic disambiguation capabilities in traditional rule / machine learning techniques. Moreover, it achieves efficient extraction without requiring large-scale labeled data, overcoming the shortcomings of traditional supervised learning models that rely heavily on labeled data and have high acquisition costs. The method also utilizes a message queue to receive aviation maintenance knowledge. This invention utilizes a maintenance text data stream and performs incremental knowledge extraction to update the initial knowledge graph, replacing the offline batch processing update mode of existing technologies. It can respond in real-time to rapidly iterating maintenance knowledge such as service announcements and airworthiness directives, solving the problem of insufficient timeliness of static knowledge graphs. By collecting user feedback data, the updated knowledge graph is continuously optimized, and feedback from intelligent question-and-answer services is used for the next optimization, forming a closed-loop iteration. This corrects potential misjudgments in large language model extraction or traditional technologies, compensates for knowledge omissions in manual extraction, and improves the completeness and reliability of the knowledge graph. Furthermore, it employs a multimodal large language model to extract structured knowledge from aviation maintenance images and / or video data and integrates it into the optimized knowledge graph, breaking the limitations of existing technologies that only focus on textual knowledge and solving the defect of missing multimodal information fusion. These embodiments of the invention can improve the efficiency, timeliness, and completeness of aviation maintenance knowledge graph construction, providing maintenance personnel with more accurate and comprehensive maintenance knowledge support, effectively ensuring flight safety and operational efficiency.

[0042] Based on the above method items, the present invention provides corresponding system items embodiments.

[0043] See Figure 3 , Figure 3 This is a structural block diagram of an aviation maintenance knowledge graph construction system provided in an embodiment of the present invention. The aviation maintenance knowledge graph construction system includes: The initial knowledge graph construction module 21 is used to extract structured knowledge from aviation maintenance texts using a large language model, and then integrate the extracted knowledge with the domain ontology to generate an initial knowledge graph. The knowledge graph update module 22 is used to receive newly input aviation maintenance text data streams through a message queue, perform incremental knowledge extraction, update the initial knowledge graph based on the incrementally extracted knowledge, and generate an updated knowledge graph. The knowledge graph optimization module 23 is used to collect user feedback data to optimize the updated knowledge graph and generate an optimized knowledge graph. The knowledge graph fusion module 24 is used to extract structured knowledge from image and / or video data in the field of aviation maintenance using a multimodal large language model, and then fuse the extracted knowledge with the optimized knowledge graph. The knowledge graph application module 25 is used to provide intelligent question answering services to users based on the fused knowledge graph; among them, the user feedback data generated during the intelligent question answering service is used to optimize the fused knowledge graph for the next time.

[0044] In one alternative embodiment, the initial knowledge graph construction module 21 is configured to: The aircraft maintenance text is segmented into text units; Constructing prompt words in the aviation domain guides a large language model to extract structured knowledge from the text units in a few-shot or zero-shot manner; wherein, the structured knowledge includes named entities, relationships between entities, and events; The extracted knowledge is disambiguated by using the large language model to identify the same entity. The disambiguated knowledge is then integrated with the aviation maintenance domain ontology to generate the initial knowledge graph.

[0045] In one optional embodiment, the knowledge graph update module 22 is configured to: Receive new input aircraft maintenance text data streams via message queue; The large language model is used to perform incremental knowledge extraction on the aviation maintenance text data stream to obtain structured knowledge fragments; Add a time attribute to the structured knowledge fragment; Structured knowledge fragments with added time attributes are fused into the initial knowledge graph through a graph database interface in an incremental update manner to generate an updated knowledge graph.

[0046] In one optional embodiment, the knowledge graph optimization module 23 is configured to: Collect user feedback data, generate reverse prompts based on the user feedback data, and train the large language model to identify erroneous knowledge extraction patterns; When preset conditions are met, the large language model is triggered to review the target knowledge entries in the updated knowledge graph that are related to the user feedback data. Based on the review results, revision suggestions are generated for the target knowledge items; The updated knowledge graph is optimized based on the proposed corrections to generate an optimized knowledge graph.

[0047] In one optional embodiment, the knowledge graph fusion module 24 is configured to: When the input is image data from the field of aircraft maintenance. A multimodal large language model is used to identify the location of components, connections, or abnormal regions in an image, generate structured knowledge describing the components, relationships, or abnormalities, and link the generated structured knowledge with entities representing the components in an optimized knowledge graph. When the input is video data from the aviation maintenance field... By using a multimodal large language model to analyze video action sequences, the structured elements of maintenance operations are identified, and a structured maintenance event chain is generated. The event chain is then stored as a standard operating procedure in an optimized knowledge graph.

[0048] In one optional embodiment, the knowledge graph application module 25 is configured to: Receive questions from users in natural language. A large language model is used as the reasoning engine to perform semantic parsing on the question, and queries and causal inferences are performed based on the fused knowledge graph to generate answers, reasoning explanations, and the knowledge sources on which the answers are based.

[0049] It should be noted that the aviation maintenance knowledge graph construction system provided in this embodiment of the invention is used to execute all the process steps of the aviation maintenance knowledge graph construction method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0050] This invention also provides a terminal device, such as... Figure 3 The diagram shown is a structural block diagram of a preferred embodiment of a terminal device provided by the present invention. The terminal device includes a processor 31, a memory 32, and a computer program stored in the memory 32 and configured to be executed by the processor 31. When the processor 31 executes the computer program, it implements the aviation maintenance knowledge graph construction method as described in any of the above embodiments.

[0051] In addition, embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the aviation maintenance knowledge graph construction method as described in any of the above embodiments.

[0052] When the processor 31 executes the computer program, it implements the steps in the above-described embodiments of the aviation maintenance knowledge graph construction method, for example... Figure 1 All steps of the aviation maintenance knowledge graph construction method shown. Alternatively, when the processor 31 executes the computer program, it implements the functions of each module in the above-described aviation maintenance knowledge graph construction system embodiment, for example... Figure 2 The functions of each module in the aviation maintenance knowledge graph construction system are shown.

[0053] Preferably, the computer program can be divided into one or more modules / units, which are stored in the memory 32 and executed by the processor 31 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device.

[0054] The processor 31 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor 31 can be any conventional processor. The processor 31 is the control center of the terminal device, connecting various parts of the terminal device through various interfaces and lines.

[0055] The memory 32 mainly includes a program storage area and a data storage area. The program storage area can store the operating system, applications required for at least one function, etc., and the data storage area can store related data, etc. In addition, the memory 32 can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, and a flash card, etc., or the memory 32 can also be other volatile solid-state storage devices.

[0056] It should be noted that the aforementioned terminal devices may include, but are not limited to, processors and memory, as will be understood by those skilled in the art. Figure 3 The structural block diagram shown is merely a structural example of the terminal device described above and does not constitute a limitation on the structure of the terminal device. The terminal device may include more or fewer components than shown, or combine certain components, or use different components.

[0057] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An aviation maintenance knowledge graph construction method, characterized in that, The method comprises the steps of: extracting structured knowledge from aviation maintenance text by using a large language model, fusing the extracted knowledge with a domain ontology, and generating an initial knowledge graph; receiving new input aviation maintenance text data stream through a message queue for incremental knowledge extraction, updating the initial knowledge graph according to the incremental knowledge extraction, and generating an updated knowledge graph; collecting user feedback data to optimize the updated knowledge graph, and generating an optimized knowledge graph; extracting structured knowledge from image and / or video data in the aviation maintenance field by using a multi-modal large language model, and fusing the extracted knowledge with the optimized knowledge graph; based on the fused knowledge graph, providing intelligent question and answer services to users; wherein the user feedback data generated in the process of the intelligent question and answer services is used to optimize the fused knowledge graph next time. 2.The method of claim 1, wherein, The method comprises the steps of: segmenting aviation maintenance text to form text units; constructing prompt words in the aviation maintenance field to guide the large language model to extract structured knowledge from the text units in a few-shot or zero-shot manner; wherein the structured knowledge includes named entities, entity relationships, and events; using the large language model to identify the same entity for the extracted knowledge to disambiguate the knowledge; fuse the disambiguated knowledge with the aviation maintenance domain ontology to generate the initial knowledge graph. 3.The method of claim 1, wherein, The method comprises the steps of: receiving new input aviation maintenance text data stream through a message queue; using the large language model to extract incremental knowledge from the aviation maintenance text data stream to obtain structured knowledge fragments; adding time attributes to the structured knowledge fragments; fuse the structured knowledge fragments with time attributes to the initial knowledge graph through a graph database interface in an incremental update manner to generate an updated knowledge graph. 4.The method of claim 1, wherein, The method comprises the steps of: collecting user feedback data, generating a reverse prompt according to the user feedback data, training the large language model to identify incorrect knowledge extraction patterns; triggering the large language model to review target knowledge items related to the user feedback data in the updated knowledge graph when the preset conditions are met; based on the review results, generate a correction suggestion for the target knowledge items; optimizing the updated knowledge graph according to the correction suggestion to generate an optimized knowledge graph. 5.The method of claim 1, wherein, When the input is image data in the aviation maintenance field, ​ The multi-modal large language model is used to identify component positions, connection relationships or abnormal areas in an image, generate structured knowledge describing the components, relationships or abnormalities, and link the generated structured knowledge to entities representing the components in the optimized knowledge graph. When the input is video data in the field of aviation maintenance, The multi-modal large language model is used to analyze the action sequence of the video, identify structured elements of the maintenance operation, and generate a structured maintenance event chain.

6. The aviation maintenance knowledge graph construction method of claim 1, wherein, Based on the fused knowledge graph, the user is provided with intelligent question and answer services, including: Receiving a question in natural language form from the user; Using a large language model as an inference engine to perform semantic analysis on the question, and performing query and causal inference based on the fused knowledge graph to generate an answer, a reasoning explanation, and the knowledge source on which the answer is based.

7. The aviation maintenance knowledge graph construction method of claim 3, wherein, The time attribute includes an effective date, an invalid date, or revision version information.

8. An aviation maintenance knowledge graph construction system, characterized in that, It includes: An initial knowledge graph construction module that uses a large language model to extract structured knowledge from aviation maintenance text, fuses the extracted knowledge with a domain ontology, and generates an initial knowledge graph; A knowledge graph updating module that receives a new input stream of aviation maintenance text data through a message queue to perform incremental knowledge extraction, updates the initial knowledge graph based on the incremental extracted knowledge, and generates an updated knowledge graph; A knowledge graph optimization module that collects user feedback data to optimize the updated knowledge graph and generates an optimized knowledge graph; A knowledge graph fusion module that uses a multi-modal large language model to extract structured knowledge from image and / or video data in the field of aviation maintenance, and fuses the extracted knowledge with the optimized knowledge graph; A knowledge graph application module that provides intelligent question and answer services to users based on the fused knowledge graph; wherein the user feedback data generated during the intelligent question and answer service process is used to optimize the fused knowledge graph for the next time.

9. A terminal device, comprising: It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor executes the computer program to implement the aviation maintenance knowledge graph construction method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium includes a stored computer program, wherein the computer readable storage medium controls the device where it is located to execute the aviation maintenance knowledge graph construction method of any one of claims 1-7 when the computer program is running.

Images