Airplane inspection method, device and storage medium

Abstract

This application provides an aircraft inspection method, device, and storage medium. The method includes: dynamically adjusting the inspection requirements of each aircraft to be inspected based on inspection auxiliary information, including historical flight data, historical inspection data, and flight environment data; acquiring inspection images of each aircraft during the inspection process based on the inspection requirements; inputting the inspection images of each aircraft into an inspection anomaly detection model to generate an inspection report for each aircraft, the inspection report indicating whether there are anomalies on the surface of each aircraft; and if the inspection report includes anomaly information, obtaining a processing decision corresponding to the anomaly information, the anomaly information indicating the presence of anomalies on the surface of the aircraft to be inspected. This method improves the accuracy and efficiency of aircraft inspection.

Description

Technical Field

[0001] This application relates to communication technology, and more particularly to an aircraft inspection method, equipment, and storage medium. Background Technology

[0002] With the rapid development of the aviation industry, aircraft safety and operational efficiency have become key considerations. Aircraft inspection, as a crucial link in ensuring flight safety, is of paramount importance in terms of accuracy and efficiency. Traditional aircraft inspection methods mainly rely on manual labor, requiring inspectors to visually examine the aircraft's surface. This method is not only time-consuming and labor-intensive but also susceptible to human factors, potentially leading to missed or incorrect inspections.

[0003] Therefore, improving the accuracy and efficiency of aircraft inspections is an urgent problem to be solved. Summary of the Invention

[0004] This application provides an aircraft inspection method, equipment, and storage medium to solve the problems of low accuracy and efficiency in aircraft inspection.

[0005] Firstly, this application provides an aircraft inspection method, including:

[0006] The inspection requirements of each aircraft to be inspected are dynamically adjusted based on the inspection auxiliary information of each aircraft to be inspected. The inspection auxiliary information includes historical flight data, historical inspection data, and flight environment data.

[0007] Based on the inspection requirements of each of the aircraft to be inspected, obtain inspection images of each of the aircraft to be inspected during the aircraft inspection process.

[0008] The inspection images of each aircraft to be inspected are input into the inspection anomaly detection model to generate an inspection report for each aircraft to be inspected. The inspection report is used to indicate whether there are any anomalies on the surface of each aircraft to be inspected.

[0009] If the inspection report includes abnormal information, then the processing decision corresponding to the abnormal information is obtained based on the abnormal information, which is used to indicate that there is an abnormality on the surface of the aircraft to be inspected.

[0010] Optionally, the step of dynamically adjusting the inspection requirements of each aircraft to be inspected based on the inspection auxiliary information of each aircraft to be inspected includes:

[0011] The inspection auxiliary information is obtained, and an inspection demand analysis network is established, wherein the inspection demand analysis network is used to determine the inspection demand of each of the aircraft to be inspected based on the inspection auxiliary information.

[0012] Anomaly trend analysis is performed on the inspection auxiliary information based on the inspection demand analysis network. Based on the historical flight data and / or historical inspection data and / or flight environment data, a set of high-risk monitoring points for the aircraft to be inspected is determined, as well as the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points.

[0013] Based on the set of high-risk monitoring points and the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points, the inspection requirements of each of the aircraft to be inspected are adjusted.

[0014] Optionally, the step of performing anomaly trend analysis on the inspection auxiliary information based on the inspection demand analysis network, determining the set of high-risk monitoring points for the aircraft to be inspected based on the historical flight data, and / or historical inspection data, and / or flight environment data, and the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points, includes:

[0015] The inspection demand analysis network determines the anomaly score calculation data for each aircraft component feature node based on the inspection auxiliary information and the mapping relationship between the inspection auxiliary information and the aircraft component feature nodes.

[0016] The correlation analysis network is used to analyze the correlation between the anomaly score calculation data and the abnormal state of the aircraft component to obtain the degree of deviation between the current state and the normal state of the characteristic node of the aircraft component, as well as the abnormal trend.

[0017] The deviation between the current state and the normal state of the aircraft component feature node, as well as the abnormal trend, are fused and calculated to generate an abnormal score for the aircraft component feature node.

[0018] Based on the abnormal scores of the aircraft component feature nodes and the abnormal score threshold, the aircraft component feature nodes that are higher than the abnormal score threshold are taken as the set of high-risk monitoring points of the aircraft to be inspected.

[0019] Based on the anomaly scores of each high-risk monitoring point in the set of high-risk monitoring points, the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points is determined.

[0020] Optionally, acquiring inspection images of each of the aircraft to be inspected during the inspection process, based on the inspection requirements of each of the aircraft to be inspected, includes:

[0021] Based on the inspection requirements of each of the aircraft to be inspected, the inspection image parameters to be collected for different aircraft component feature nodes of each of the aircraft to be inspected are determined. The inspection image parameters include the number of inspection images, and / or the angle of the inspection images, and / or the clarity of the inspection images.

[0022] The intelligent image capture model extracts the acquisition features of the inspection image parameters, identifies and classifies the image features, and generates image acquisition parameters for acquiring inspection images of each of the aircraft to be inspected.

[0023] The inspection images of each of the aircraft to be inspected are acquired according to the image acquisition parameters.

[0024] Optionally, the anomaly detection model includes a visual recognition model and a deep learning model. The step of inputting the inspection images of each aircraft to be inspected into the anomaly detection model to generate inspection reports for each aircraft to be inspected includes:

[0025] The visual recognition model is used to extract image features from the inspection images. The image features include the shape, texture, and color of the aircraft parts of the aircraft to be inspected.

[0026] The image features in the inspection image are mapped to the high-dimensional space of the deep learning model through a trained deep learning model to generate the feature vector corresponding to the image features.

[0027] Based on the similarity calculation results between the feature vector corresponding to the image feature and the standard feature vector, the abnormal features of each aircraft component feature node of the aircraft to be inspected are determined, and the abnormal state and / or abnormal position corresponding to the abnormal features are marked.

[0028] Based on the abnormal characteristics of the characteristic nodes of each aircraft component of each aircraft to be inspected, an inspection report is generated for each aircraft to be inspected.

[0029] Optionally, obtaining the processing decision corresponding to the anomaly information based on the anomaly information includes:

[0030] Obtain the first knowledge graph corresponding to the abnormal information, and the second knowledge graph corresponding to the processing decision;

[0031] Based on the association path of the abnormal information in the first knowledge graph, the associated abnormal information is determined, and the associated abnormal information is the abnormal information corresponding to the first abnormal entity that has an abnormal association with the abnormal information.

[0032] Based on the correlation between the second abnormal entity corresponding to the abnormal information, the first abnormal entity, and the visualization data matrix of the first knowledge graph and the visualization data matrix of the second knowledge graph, at least one processing decision entity in the second knowledge graph corresponding to the first abnormal entity and the second abnormal entity is determined.

[0033] Based on the processing decision corresponding to the processing decision entity, determine the processing decision corresponding to the abnormal information;

[0034] The steps of obtaining the first knowledge graph corresponding to the abnormal information and the second knowledge graph corresponding to the processing decision include:

[0035] An aviation knowledge base is pre-established, which contains knowledge data on aircraft components, common faults, maintenance procedures, and safety standards. The aviation knowledge base is generated by organizing and compiling technical documents from aircraft manufacturers, professional books in the aviation field, historical inspection records, and maintenance reports.

[0036] Using natural language processing algorithms, entity recognition and relation extraction are performed on the abnormal information to determine the key entity nouns in the abnormal information and the relationships between the key entity nouns, and a first processing result is generated.

[0037] Based on the first processing result, a first knowledge graph is constructed. In the first knowledge graph, each node represents a key entity noun, and the edges represent the relationships between the key entity nouns.

[0038] In addition, decision data related to the processing decision is acquired, and the decision data is derived from maintenance manuals, best practice guidelines, and historical processing cases;

[0039] The decision data is subjected to entity recognition and relation extraction using natural language processing algorithms to generate a second processing result.

[0040] Based on the second processing result, a second knowledge graph is constructed. In the second knowledge graph, each node represents a processing strategy, maintenance tool or spare part, and the edge represents the logical relationship or dependency relationship between nodes.

[0041] After generating the first knowledge graph and the second knowledge graph, the association between the first knowledge graph and the second knowledge graph is established, specifically by adding cross-knowledge graph edges to the first knowledge graph and the second knowledge graph;

[0042] The step of determining the associated abnormal information based on the association path of the abnormal information in the first knowledge graph includes:

[0043] By matching keywords in the abnormal information, the target node corresponding to the abnormal information is located in the first knowledge graph. The target node is used as the starting point, and potential related paths are explored along the edges starting from the starting point node. In the first knowledge graph, nodes are connected by the edges, which represent the relationship between nodes.

[0044] During the exploration process, each node and edge visited is recorded, as well as the degree of association between each node and edge and the starting node. The degree of association is represented by weights.

[0045] After exploring all potential related paths, analyze and record all candidate nodes that are directly or indirectly connected to the starting node. The candidate nodes represent other abnormal information associated with the abnormal information.

[0046] The association anomaly information is sorted and filtered according to the degree of association and the path length of the potential association path to generate filtered association anomaly information;

[0047] The step of locating the target node corresponding to the abnormal information in the first knowledge graph by matching keywords in the abnormal information includes:

[0048] Keywords are extracted from the abnormal information, and the keywords include at least the aircraft component name, the fault type, and the abnormal phenomenon.

[0049] After extracting the keywords, the keywords are preprocessed, including stop word removal and stemming.

[0050] A word embedding algorithm is used to map the preprocessed keywords into a high-dimensional vector space to generate keyword vectors;

[0051] The first knowledge graph is traversed using an approximate nearest neighbor search algorithm, and the cosine similarity between each node and the keyword vector is calculated.

[0052] If the cosine similarity between any node and the keyword vector is greater than a set threshold, the node is determined as the target node corresponding to the abnormal information.

[0053] Furthermore, if multiple nodes have a cosine similarity greater than a set threshold, then the nodes are sorted according to the magnitude of the cosine similarity, and the node with the highest similarity is selected as the target node.

[0054] The step of taking the target node as the starting point and exploring potential associated paths along the edges from the starting node includes:

[0055] Initialize the exploration environment, which specifically includes loading the complete graph data of the first knowledge graph and setting the corresponding exploration parameters, including the maximum exploration depth and the step size of each exploration.

[0056] A breadth-first search algorithm is used to start from the target node and explore along the edges connected to it. Every node and edge visited is recorded, as well as the degree of association between each node and edge and the starting node. During the exploration process, associated paths that meet the preset mining value are selected based on the set evaluation criteria. The evaluation criteria include evaluation criteria based on node attributes, edge type or weight.

[0057] The discovered connections will be visualized.

[0058] Optionally, determining the processing decision corresponding to the abnormal information based on the processing decision corresponding to the processing decision entity includes:

[0059] Obtain the maintenance history, spare parts inventory, and maintenance support information of the aircraft to be inspected;

[0060] The processing decision is adjusted based on the maintenance history information, the spare parts inventory, the repair support information, and the processing decision corresponding to the processing decision entity.

[0061] The adjusted processing decision is determined as the processing decision corresponding to the abnormal information.

[0062] Optionally, the method further includes:

[0063] If there is abnormal feedback information for the processing decision corresponding to the abnormal information, and a call request is received from the inspection terminal, then according to the call request and the status of the maintenance terminal, the idle maintenance terminal is called. The call request is used to request to communicate with the maintenance terminal to provide abnormal feedback.

[0064] If the idle maintenance terminal accepts the call request, a call connection is established between the inspection terminal and the idle maintenance terminal;

[0065] If the idle maintenance terminal does not respond to the call request, it will call other idle maintenance terminals until a call connection is established between the inspection terminal and the other idle maintenance terminals.

[0066] Secondly, this application provides an electronic device, including: a processor, a communication interface, and a memory, wherein the processor is communicatively connected to the communication interface and the memory respectively;

[0067] The memory stores computer-executed instructions;

[0068] The communication interface communicates and interacts with external devices.

[0069] The processor executes computer execution instructions stored in the memory to implement the method as described in any one of the first aspects.

[0070] Thirdly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the aircraft inspection method as described in any one of the first aspects.

[0071] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the aircraft inspection method described in any one of the first aspects.

[0072] The aircraft inspection method, equipment, and storage medium provided in this application dynamically adjust the inspection requirements of each aircraft to be inspected based on the inspection auxiliary information of each aircraft. Based on the inspection requirements of each aircraft, inspection images of each aircraft are acquired during the inspection process. These images are then input into an inspection anomaly detection model to generate inspection reports for each aircraft. If the inspection report includes anomaly information, a corresponding processing decision is obtained based on the anomaly information. This method effectively reduces manual intervention in the aircraft inspection process, lowers the workload, and improves the efficiency of the inspection process. Furthermore, by using image recognition methods, anomalies in the aircraft inspection images are automatically identified, and corresponding processing decisions are provided for each anomaly, further improving the accuracy of aircraft inspection and anomaly handling. Attached Figure Description

[0073] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0074] Figure 1 A flowchart illustrating an aircraft inspection method provided in this application embodiment;

[0075] Figure 2 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0076] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0077] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0078] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems will be described in detail below with reference to specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0079] Figure 1 This is a flowchart illustrating an aircraft inspection method provided in an embodiment of this application. Figure 1 As shown, the method may include:

[0080] S101. Dynamically adjust the inspection requirements of each aircraft to be inspected based on the inspection auxiliary information of each aircraft to be inspected.

[0081] The inspection auxiliary information refers to various types of information that provide decision support to the server during aircraft inspections, including but not limited to historical flight data, historical inspection data, and flight environment data. Historical flight data includes historical records such as flight duration, route, altitude, and speed. This data reflects the damage to the aircraft fuselage during past flights; for example, specific routes may cause damage to specific parts of the aircraft. In some cases, it can also reflect the predicted degree of cloud damage. Historical inspection data refers to past inspection records, including inspection time, results, and problems found. Combining historical inspection data helps identify high-risk areas of the aircraft, allowing for more targeted inspection plans for those areas. Flight environment data includes information about the external environment in which the aircraft operates, such as weather conditions and air quality. In some cases, this flight environment data can also reflect the extent of fuselage damage.

[0082] The server can flexibly modify inspection plans based on real-time collected information to meet the inspection needs of different aircraft and under different conditions. Specifically, it collects and integrates historical flight data, historical inspection data, and flight environment data. Analyzing this data identifies potential problem areas and risk points. Based on the analysis results, the inspection focus, frequency, and methods for each aircraft to be inspected are adjusted.

[0083] For example, if the server receives data for five aircraft to be inspected, and by analyzing historical flight data, it finds that two of these aircraft have recently flown for extended periods and frequently fly in adverse weather conditions, and that historical inspection data shows minor damage to the wings of these two aircraft, the server can increase the frequency and accuracy of inspections of the wings of these two aircraft based on this information.

[0084] S102. Based on the inspection requirements of each of the aircraft to be inspected, obtain inspection images of each of the aircraft to be inspected during the aircraft inspection process.

[0085] Among them, inspection images refer to high-resolution images of various parts of the aircraft taken during the inspection process, used for subsequent analysis and judgment. These inspection images can be taken using a high-definition camera or other image acquisition equipment, with the shooting location, angle, and resolution determined according to the inspection requirements.

[0086] In one possible implementation, inspection requirements can include standardized inspections and flexible inspections. Standardized inspections refer to inspections of a specific model with designated inspection areas, and mandatory photography for certain key inspection areas. After each area is inspected, an image of that area is uploaded and marked as inspected; the inspection only ends when all areas have been inspected. Flexible inspections, on the other hand, do not restrict specific inspection areas or photography requirements; inspectors can determine the inspection process and whether to take photos independently.

[0087] Under standardized inspection procedures, the inspection images can be taken manually by inspectors, or the server can determine the shooting location, angle, and resolution of the images based on inspection requirements and send these parameters to devices such as drones, where the image acquisition devices automatically capture the images. After the inspection images are captured, they can be uploaded to the server as part of the inspection record for maintenance personnel to review or for subsequent quality inspection checks.

[0088] Optionally, when inspection images are taken manually by inspection personnel, if serious problems exist in areas outside the areas covered by the inspection requirements, the inspection personnel can take additional inspection images of that area and upload them to the server as emergency inspection images. Uploaded emergency inspection images can be displayed with higher priority in the maintenance monitoring list, allowing them to receive processing recommendations at a higher rate.

[0089] S103. Input the inspection images of each aircraft to be inspected into the inspection anomaly detection model to generate an inspection report for each aircraft to be inspected.

[0090] The inspection report is used to indicate whether there are any anomalies on the surface of each aircraft to be inspected. The anomaly detection model is a machine learning-based model capable of automatically identifying anomalies in the inspection images. The inspection report is a document that records the inspection results for each aircraft in detail, including information such as whether anomalies exist, the location and type of the anomalies.

[0091] The inspection images are input into the inspection anomaly detection model for processing. The model automatically identifies aircraft components in the images and detects whether any anomalies exist. An inspection report containing the inspection results is generated. For example, the server can input the inspection image obtained from the aforementioned S102 step into a pre-trained inspection anomaly detection model. The model performs depth analysis on the image, automatically identifies a tiny crack in the wing area, and records the location, size, and shape of the crack in detail in the inspection report.

[0092] S104. If the inspection report includes abnormal information, then obtain the processing decision corresponding to the abnormal information based on the abnormal information.

[0093] The anomaly information indicates anomalies on the surface of the aircraft to be inspected. This anomaly information may include anomalies recorded in the inspection report, such as cracks or corrosion. The processing decision is a solution or maintenance recommendation based on the anomaly information.

[0094] Based on the inspection report, the server determines the location and condition of any anomalies on the inspected aircraft. Then, it queries the database for corresponding handling decisions and outputs these decisions to maintenance personnel. The maintenance personnel then promptly repair the aircraft according to these decisions. For example, the server might discover a tiny crack in the wing area based on the inspection report. The server then queries its database for handling decisions based on the crack's size and location to determine the corresponding repair plan. Subsequently, the server sends the repair plan to the maintenance personnel and guides them in promptly repairing the crack.

[0095] The method provided in this application dynamically adjusts the inspection requirements of each aircraft to be inspected based on the inspection auxiliary information of each aircraft to be inspected. Based on the inspection requirements of each aircraft to be inspected, inspection images of each aircraft to be inspected are acquired during the inspection process. These images are then input into an inspection anomaly detection model to generate inspection reports for each aircraft. If the inspection report includes anomaly information, a corresponding processing decision is obtained based on the anomaly information. This method effectively reduces manual intervention in the aircraft inspection process, lowers the workload, and improves the efficiency of the aircraft inspection process. Furthermore, by using image recognition methods, anomalies in the aircraft inspection images are automatically identified, and corresponding processing decisions are provided for each anomaly, further improving the accuracy of aircraft inspection and anomaly handling.

[0096] In one possible implementation, the aforementioned step S101 may include:

[0097] S1011. Obtain the inspection auxiliary information and the inspection demand analysis network.

[0098] The inspection demand analysis network is used to determine the inspection requirements of each of the aircraft to be inspected based on the inspection auxiliary information.

[0099] Among them, the inspection requirements analysis network is a neural network model that, through training and learning, can intelligently determine the specific inspection requirements of each aircraft to be inspected based on the input inspection auxiliary information.

[0100] The server can extract or receive inspection auxiliary information from the database, including historical flight data (such as flight duration, route, etc.), historical inspection data (past inspection records and problems found), and flight environment data (such as weather conditions and specific environmental factors of the flight area). It can then load or access a pre-trained inspection requirements analysis network model to prepare for subsequent data analysis.

[0101] For example, the server retrieves relevant data for five aircraft to be inspected from its internal database, including their historical flight records, past inspection reports, and recent flight environment information. Simultaneously, the server loads a trained inspection requirements analysis network that can predict potential problems and inspection priorities for various parts of the aircraft based on this data.

[0102] S1012. Perform anomaly trend analysis on the inspection auxiliary information according to the inspection demand analysis network, determine the set of high-risk monitoring points of the aircraft to be inspected based on the historical flight data, and / or historical inspection data, and / or flight environment data, and determine the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points.

[0103] The server can utilize the inspection demand analysis network to perform in-depth analysis of the collected inspection auxiliary information, predicting potential anomalies or failure trends in various aircraft components. Based on these potential anomalies or failure trends, the system identifies a set of aircraft components or areas requiring focused attention and inspection during the inspection process. For each point in the high-risk monitoring point set, a priority is assigned based on its potential risk level and urgency to guide the inspection sequence and focus.

[0104] For example, the server can input the collected data into the inspection requirements analysis network. After analysis, the network outputs high-risk monitoring points for each aircraft that require special attention, such as a specific area on the wing or the engine connection point. At the same time, the network also assigns different inspection priorities to these monitoring points. For example, a certain area on the wing, due to multiple historical damage records, is given the highest inspection priority.

[0105] Specifically, step S1012 may include:

[0106] S10121. The inspection demand analysis network can determine the anomaly score calculation data of each aircraft component feature node based on the inspection auxiliary information and the mapping relationship between the inspection auxiliary information and the aircraft component feature nodes.

[0107] The aircraft component feature nodes refer to the characteristic points of various key components or structures of the aircraft, which reflect the status and performance of the components. The anomaly score calculation data is derived based on the mapping relationship between inspection auxiliary information and aircraft component feature nodes, resulting in data used to calculate the anomaly score.

[0108] The server establishes a mapping relationship between the acquired inspection auxiliary information and the feature nodes of aircraft components. Using this data and mapping relationship, the server extracts anomaly scoring data related to each aircraft component feature node. For example, for the aircraft wing component, the server extracts wind speed and direction information from historical flight data, wing damage records from historical inspection data, and temperature and humidity information from flight environment data, as the basis for calculating the wing component's anomaly score.

[0109] S10122. By performing correlation analysis on the abnormal score calculation data and the abnormal state of the aircraft component through a correlation analysis network, the degree of deviation between the current state and the normal state of the characteristic node of the aircraft component, as well as the abnormal trend, are obtained.

[0110] The correlation analysis network can be a neural network model used to analyze the correlation between the anomaly score calculation data and the abnormal state of the aircraft component. The degree of deviation indicates the difference between the current state and the normal state of the aircraft component. The anomaly trend is the predicted trend of component performance changes based on historical and current data.

[0111] The server inputs the extracted anomaly score calculation data into a correlation analysis network. This network performs correlation analysis between this data and the abnormal states of aircraft components. Taking wing components as an example, the network analyzes the correlation between environmental factors such as wind speed, wind direction, temperature, and humidity and wing damage, calculates the degree of deviation between the current state and the normal state of the wing component, and predicts abnormal trends. For instance, the network may find a high correlation between high wind speed and specific wind direction and wing damage, thereby determining the degree of deviation of the wing component in the current environment and potential abnormal trends.

[0112] S10123. The deviation between the current state and the normal state of the aircraft component feature node and the abnormal trend are fused and calculated to generate an abnormal score for the aircraft component feature node.

[0113] The server generates an anomaly score for each aircraft component feature node based on the fusion calculation results of the deviation degree and the anomaly trend. This score reflects the abnormal risks and potential problems of the component in the current environment. If the deviation degree of the aircraft component is large and the anomaly trend is obvious, then its anomaly score will be high.

[0114] For example, the server uses a weighted average algorithm to perform a fusion calculation based on the deviation degree and abnormal trend of the wing feature nodes, resulting in a specific anomaly score (e.g., 75 points out of 100, with higher scores indicating better conditions). This score intuitively reflects the current degree of anomaly and potential risks of the wing.

[0115] S10124. Based on the abnormal scores of the aircraft component feature nodes and the abnormal score threshold, the aircraft component feature nodes that are higher than the abnormal score threshold are taken as the set of high-risk monitoring points of the aircraft to be inspected.

[0116] The anomaly scoring threshold is a preset score used to determine whether an aircraft component is in a high-risk state. The server compares the anomaly scores of each aircraft component's feature nodes with the preset anomaly scoring threshold. Component nodes with scores exceeding the threshold are identified as high-risk monitoring points and added to the high-risk monitoring point set. For example, if the anomaly score of a wing component exceeds the threshold, it will be added to the high-risk monitoring point set.

[0117] For example, the anomaly score (75 points) of the wing feature node is compared with the preset anomaly score threshold (80 points). If the score is found to be lower than the threshold, the wing feature node can be added to the set of high-risk monitoring points.

[0118] S10125. Based on the anomaly scores of each high-risk monitoring point in the set of high-risk monitoring points, determine the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points.

[0119] Within the set of high-risk monitoring points, the server can determine their inspection priorities based on their anomaly scores. Component nodes with higher anomaly scores receive higher inspection priorities. For example, if a wing component has the highest anomaly score among all high-risk monitoring points, it will be assigned the highest inspection priority to ensure it is inspected and addressed first during the inspection process.

[0120] For example, in the set of high-risk monitoring points, besides the wing feature node, there is also an engine feature node with an anomaly score of 70. The server sorts these two nodes based on their anomaly scores, determining that the wing feature node has a lower inspection priority than the engine feature node, because the engine feature node's lower anomaly score indicates a worse condition or higher risk. Therefore, the engine feature node will be inspected first during the inspection process.

[0121] S1013. Based on the set of high-risk monitoring points and the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points, adjust the inspection requirements of each of the aircraft to be inspected.

[0122] After the server has completed the analysis of the inspection auxiliary information of the aircraft to be inspected and determined the set of high-risk monitoring points for each aircraft and the inspection priority of these monitoring points, it needs to adjust the inspection plan for each aircraft based on this information.

[0123] Taking a specific aircraft to be inspected as an example, suppose the server, through analysis, discovers that the wing joint and engine air intake are two high-risk monitoring points for this aircraft model, and that the inspection priority of the wing joint is higher than that of the engine air intake. Based on these findings, the server can make at least one of the following adjustments:

[0124] Adjusting image acquisition priorities: For wing joints, the server can instruct inspection equipment (such as drones or robots) or personnel to increase the frequency and clarity of image capture of this area. For example, a drone can be programmed to capture multi-angle, high-resolution images of the wing joint during inspection. While engine blades have a relatively lower inspection priority, image coverage and clarity still need to be ensured, although the frequency and angle of image capture can be slightly less compared to the wing joint.

[0125] Optimize image processing algorithms: The server can adjust or optimize image processing algorithms to more accurately identify and analyze anomalies at wing joints and engine blades. This may include enhancing algorithms such as edge detection, color analysis, or shape recognition to improve the accuracy of automatically identifying potential problems.

[0126] Customized image recognition model: Based on historical data and current high-risk monitoring points, the server may train or adjust a specialized image recognition model to more accurately identify cracks, corrosion or other damage on wing joints and engine blades.

[0127] Setting image recognition priority: When performing automatic image recognition, the server will process the image at the wing connection first to ensure that high-priority monitoring points can receive faster analysis and feedback.

[0128] The method provided in this application utilizes an inspection demand analysis network to perform in-depth analysis and processing of inspection auxiliary information, intelligently determining the specific inspection needs of each aircraft to be inspected. This enables personalized inspection plan development, optimizes inspection resource allocation, and reduces the blind spots and repetitiveness in the inspection process. Through anomaly trend analysis, this method accurately identifies the set of high-risk monitoring points for the aircraft to be inspected and the inspection priorities of these points. This allows inspection work to focus more on areas with high potential risks, promptly identifying and addressing potential problems, thereby improving the safety and reliability of aircraft operations.

[0129] In one possible implementation, step S102 may include:

[0130] S1021. Based on the inspection requirements of each of the aircraft to be inspected, determine the inspection image parameters that need to be collected for different aircraft component feature nodes of each of the aircraft to be inspected.

[0131] The inspection image parameters include the number of inspection images, and / or the angle of the inspection images, and / or the clarity of the inspection images.

[0132] The server can analyze the inspection requirements for each aircraft to be inspected. These requirements may include special attention to characteristic nodes of different aircraft components, such as wing joints and engine air intakes. Based on these requirements, the server determines the specific parameters for acquiring inspection images of these characteristic nodes.

[0133] For example, for the wing joint, due to its complex structure and vulnerability, the server may determine that multiple high-resolution images need to be captured to show the details of the joint from different angles. Therefore, the inspection image parameters for this feature node may include: 10 images, taken from different angles (such as front, side, oblique angle, etc.), and each image must have a resolution of 1080P or higher.

[0134] S1022. Extract the acquisition features of the inspection image parameters through the intelligent image capture model, and identify and classify the image features to generate image acquisition parameters for acquiring inspection images of each of the aircraft to be inspected.

[0135] The intelligent image capture model is a machine learning-based model capable of analyzing and extracting image features to optimize the image acquisition process. These image features are key information reflecting the content and attributes of the image, such as edges, textures, and colors. The image acquisition parameters include the number of inspection images set to meet inspection requirements, and / or the angle and / or sharpness of the inspection images.

[0136] The server can utilize an intelligent image capture model to further refine the acquisition parameters of the inspection images. This model can identify the unique acquisition features of different aircraft component feature nodes, such as specific shapes, textures, or colors. Taking the wing joint as an example, the intelligent image capture model may identify key components such as bolts and sealing strips at the joint, and determine the optimal shooting angle and lighting conditions based on the characteristics of these components. In this way, the generated image acquisition parameters will be more accurate, ensuring that the acquired inspection images can clearly show the condition of these key components.

[0137] S1023. Collect inspection images of each of the aircraft to be inspected according to the image acquisition parameters.

[0138] Based on the precise image acquisition parameters generated by the intelligent image capture model, the server can guide inspection equipment (such as drones or inspection robots) or inspection personnel to acquire inspection images. During this process, the server ensures that the acquired images strictly conform to the previously determined parameter requirements, such as quantity, angle, and sharpness.

[0139] Taking server-guided drone inspection equipment as an example, the server can instruct the drone to take pictures at predetermined angles and distances, ensuring that each image clearly shows the details of the connection points. Simultaneously, the server will perform real-time quality checks on the acquired images to ensure that the image quality meets the needs of subsequent analysis.

[0140] Taking server-guided inspection personnel as an example, the server can instruct the personnel to use image acquisition equipment equipped with a high-resolution lens to capture images of various feature nodes of the wing according to the image acquisition parameters provided by the intelligent image capture model. After the images are captured, the personnel can immediately perform a quality check on the images to ensure that the captured images meet the requirements of the image acquisition parameters.

[0141] The method provided in this application embodiment determines the required inspection image parameters for different aircraft component feature nodes of each aircraft to be inspected based on the inspection requirements of each aircraft to be inspected. It then extracts the acquisition features of the inspection image parameters using an intelligent image capture model, identifies and classifies these image features, generates image acquisition parameters for acquiring inspection images of each aircraft to be inspected, and acquires inspection images of each aircraft to be inspected based on these image acquisition parameters. This achieves adaptive determination of the acquisition parameters for acquiring inspection images of the aircraft to be inspected according to their inspection requirements, improving the efficiency and accuracy of inspection image acquisition.

[0142] In one possible implementation, the anomaly detection model includes a visual recognition model and a deep learning model, and the aforementioned step S103 may include:

[0143] S1031. Extract image features from the inspection images using a visual recognition model.

[0144] The image features include the shape, texture, and color of the aircraft components of the aircraft to be inspected. The visual recognition model can be any existing algorithm for visual recognition, used to automatically extract key information from the inspection image. In this application, the key information includes information to determine whether the aircraft to be inspected has any inspection anomalies. For example, the visual recognition model can extract key information such as whether the shape of the aircraft components has been deformed, whether the color of the aircraft components has changed, and whether the texture of the aircraft components is abnormal, to determine whether the aircraft components of the aircraft to be inspected may have damage or other anomalies.

[0145] By extracting features from the inspection images using a visual recognition model, image features such as the shape, texture, and color of the aircraft components to be inspected can be obtained. Based on these image features, it can be determined whether the aircraft components meet the standards, thus confirming whether the components are functioning properly. For example, whether the shape of the wing conforms to the standard profile, whether the texture of the engine blades is regular, and whether the paint color on the aircraft surface is uniform, etc., can provide a data foundation for subsequent anomaly detection.

[0146] S1032. The image features in the inspection image are mapped to the high-dimensional space of the deep learning model through the trained deep learning model to generate the feature vector corresponding to the image features.

[0147] After extracting image features, the server can input these features into a pre-trained deep learning model. This model maps these features into a high-dimensional space, generating a unique feature vector for each feature. In this high-dimensional space, these feature vectors more clearly reveal subtle differences between components, helping the server to more accurately identify anomalies. Generating feature vectors corresponding to the image features in a high-dimensional space using a deep learning model improves the sensitivity of anomaly detection and provides richer and more detailed information for subsequent data analysis.

[0148] S1033. Based on the similarity calculation results between the feature vector corresponding to the image feature and the standard feature vector, determine the abnormal features of each aircraft component feature node of the aircraft to be inspected, and mark the abnormal state and / or abnormal position corresponding to the abnormal features.

[0149] The standard feature vector represents the feature vector of an aircraft component under normal conditions, and this standard feature vector can be obtained through training with a large number of positive samples.

[0150] After generating feature vectors, the server compares these vectors with pre-stored standard feature vectors. These standard feature vectors are constructed based on the features of normal aircraft components, representing the "standard appearance" of aircraft components under normal operating conditions. By calculating the similarity between the feature vector to be detected and the standard feature vectors, the server can determine which components may have anomalies. For example, if the color feature vector of a component differs significantly from the standard vector, then this component may have problems such as paint peeling or corrosion. The server accurately marks these abnormal features and indicates their corresponding abnormal states and locations, so that maintenance personnel can quickly locate and handle the problems.

[0151] For example, if the similarity between a feature vector and a standard feature vector is lower than a set similarity threshold, it can be determined that the aircraft component corresponding to that feature vector is abnormal, and the abnormal state and / or abnormal location can be marked. For instance, if the texture features of a certain area of ​​the wing differ significantly from the standard features, then that area may be marked as abnormal.

[0152] S1034. Based on the abnormal characteristics of the characteristic nodes of each aircraft component of each aircraft to be inspected, generate an inspection report for each aircraft to be inspected.

[0153] After completing anomaly detection on all aircraft component feature nodes, the server generates an inspection report based on the results. This report details all tagged anomalies on each aircraft to be inspected, including the type, location, and severity of the anomalies. Furthermore, the report can be prioritized based on the urgency and impact of anomalies, ensuring maintenance personnel can address those issues with the greatest impact on flight safety first. This inspection report not only provides clear guidance for maintenance personnel but also significantly improves the efficiency and accuracy of aircraft maintenance. Through this report, airlines can also gain a better understanding of the overall aircraft condition, enabling them to develop more appropriate maintenance plans.

[0154] The method provided in this application extracts image features from the inspection image using a visual recognition model, maps the image features in the inspection image to the high-dimensional space of the deep learning model using a trained deep learning model to generate feature vectors corresponding to the image features, determines the abnormal features of the feature nodes of each aircraft component of the aircraft to be inspected based on the similarity calculation results between the feature vectors corresponding to the image features and the standard feature vectors, marks the abnormal states and / or abnormal positions corresponding to the abnormal features, and generates an inspection report for each aircraft to be inspected based on the abnormal features of the feature nodes of each aircraft component of each aircraft to be inspected. This achieves efficient, accurate, and intelligent processing of the aircraft inspection process, improving the efficiency and safety of aircraft inspection.

[0155] In one possible implementation, step S104 may include:

[0156] S1041. Obtain the first knowledge graph corresponding to the abnormal information, and the second knowledge graph corresponding to the processing decision.

[0157] The first knowledge graph and / or the second knowledge graph may be obtained from a pre-built knowledge base. For example, it may be obtained from a pre-built aviation knowledge base, which contains knowledge data corresponding to aircraft components, common faults, maintenance procedures, and safety standards. The aviation knowledge base can be generated by organizing and compiling technical documents from aircraft manufacturers, professional books in the aviation field, historical inspection records, and maintenance reports.

[0158] Specifically, taking the first knowledge graph and / or the second knowledge graph as examples obtained from a pre-built aerospace domain knowledge base, step 1041 may specifically include:

[0159] Using natural language processing algorithms, entity recognition and relation extraction are performed on the abnormal information to determine key entity nouns and the relationships between them, generating a first processing result. Key entity nouns may include, for example, names of aircraft parts such as wings and nose cones, and fault types such as cracks and dents. The relationships between key entity nouns may be, for example, associations between aircraft component names and fault types.

[0160] Based on the first processing result, a first knowledge graph is constructed. In the first knowledge graph, each node represents a key entity noun, and edges represent the relationships between the key entity nouns. For example, in the first knowledge graph, each node can represent a key entity noun, such as "wing" or "crack". The edges between nodes can represent the fault types corresponding to the aircraft parts; for example, the edge between the wing node and the crack node represents the presence of a crack on the wing.

[0161] In one possible example, the entity recognition result of the natural language processing algorithm on the abnormal information may include three entity nouns: airplane, left wing, and crack. The relation extraction result of the natural language processing algorithm on the abnormal information may include, for example, "airplane-contains-left wing" and "left wing-appears-crack". Correspondingly, in this first knowledge graph, nodes include "airplane", "left wing", and "crack", and edges represent the relationship between each pair of nodes (e.g., the edge between airplane and left wing represents an containment relationship).

[0162] Further, decision data related to the processing decision is acquired, and this decision data originates from maintenance manuals, best practice guidelines, historical processing cases, etc. Natural language processing algorithms are used to perform entity recognition and relation extraction on the decision data to generate a second processing result. Based on the second processing result, a second knowledge graph is constructed. In the second knowledge graph, each node represents a processing strategy, maintenance tool, or spare part, and edges represent logical relationships or dependencies between nodes.

[0163] In this context, the entity nouns in the decision data serve as nodes in the second knowledge graph. These nouns can include terms such as processing strategies, repair tools, or spare parts. For example, the processing strategy could be "left wing crack repair," the repair tool could be "repair tool," and the material could be "crack repair material."

[0164] The relationships between these entity nouns are represented as edges between nodes in the second knowledge graph. For example, the extracted relationship results could include: Left wing crack repair - requires - repair tool, Left wing crack repair - requires - crack repair material.

[0165] After generating the first knowledge graph and the second knowledge graph, an association between the first knowledge graph and the second knowledge graph is established, specifically by adding cross-knowledge-graph edges to the first knowledge graph and the second knowledge graph.

[0166] Exemplarily, continuing with the above example, for instance, an edge that crosses the first knowledge graph and the second knowledge graph can be added between the "crack" node and the "left wing crack repair" node, indicating that when a left wing crack is detected, the processing decision of left wing crack repair should be taken.

[0167] S1042. Determine the associated abnormal information of the abnormal information according to the association path of the abnormal information in the first knowledge graph. The associated abnormal information is the abnormal information corresponding to the first abnormal entity that has an abnormal association with the abnormal information.

[0168] Exemplarily, after obtaining the first knowledge graph, the server can analyze the associated abnormal information of the wing crack. By traversing the association paths in the graph, the server discovers the first abnormal entity - "wing structural fatigue" that has an abnormal association with the wing crack. This is because wing cracks are often a direct manifestation of structural fatigue. By further exploring the abnormal information corresponding to "wing structural fatigue", it is determined that this fatigue may be caused by long-term flight stress, environmental factors, or improper maintenance. These information provide important background for subsequent processing decisions.

[0169] Specifically, the target node corresponding to the abnormal information can be located in the first knowledge graph by matching the keywords in the abnormal information. This step can be achieved by the following method:

[0170] Extract keywords from the abnormal information. The keywords at least include aircraft component names, fault types, abnormal phenomena, etc. Precisely extract the keywords from the abnormal information. These keywords mainly include aircraft component names (such as wings, engines, etc.), fault types (such as cracks, fractures, etc.), and abnormal phenomena (such as oil leakage, overheating, etc.). The purpose of this step is to screen out the most critical and representative words from a large amount of text information for subsequent precise matching in the knowledge graph.

[0171] After extracting the keywords, preprocess the keywords. The preprocessing includes removing stop words and stemming. Among them, stop words refer to words that frequently appear in the text but contribute little to the meaning of the text, such as "de", "le", etc. These words will generate noise during the search process and thus need to be removed. Stemming is to restore the vocabulary to its basic form, for example, restoring "flying" to "fly", which can improve the accuracy and efficiency of the search.

[0172] Word embedding algorithms are used to map preprocessed keywords into a high-dimensional vector space, generating keyword vectors. For example, word embedding algorithms (such as Word2Vec and GloVe) can be used to map preprocessed keywords into a high-dimensional vector space, generating corresponding keyword vectors. The keyword vectors generated by word embedding algorithms can capture the semantic relationships between words, making semantically similar words closer together in the vector space.

[0173] The first knowledge graph is traversed using an approximate nearest neighbor search algorithm, and the cosine similarity between each node and the keyword vector is calculated. Cosine similarity is an indicator that measures the similarity between two vectors in a direction; its value ranges from -1 to 1, with values ​​closer to 1 indicating greater similarity. For example, this can be achieved by traversing all nodes in the first knowledge graph and calculating the cosine similarity between each node and the keyword vector during the traversal. If the cosine similarity between any node and the keyword vector is greater than a set threshold, that node is identified as the target node corresponding to the abnormal information.

[0174] Optionally, if multiple nodes have a cosine similarity greater than a set threshold, then the nodes are sorted according to the magnitude of the cosine similarity, and the node with the highest similarity is selected as the target node.

[0175] After locating the target node corresponding to the anomaly information, the target node can be used as a starting point to explore potential related paths along the edges. This method can be implemented as follows:

[0176] Initialize the exploration environment, which specifically includes loading the complete first knowledge graph data, including all nodes (such as aircraft parts, fault types, etc.) and edges (representing the relationships between nodes), ensuring that all data is correctly loaded into memory for subsequent exploration operations.

[0177] Set corresponding exploration parameters, including the maximum exploration depth (i.e., the maximum number of nodes that can be traversed from a single node to find related paths) and the step size for each exploration (the number of nodes in the next level that can be traversed from the current node during one exploration). Setting these parameters helps control the scope and accuracy of the exploration.

[0178] A breadth-first search algorithm is used to start from the target node and explore along the edges connected to it. Every node and edge visited is recorded, as well as the degree of association between each node and edge and the starting node. During the exploration process, associated paths that meet the preset mining value are selected based on the set evaluation criteria, which include evaluation criteria based on node attributes, edge types, or weights.

[0179] Specifically, starting from the target node, a breadth-first search algorithm is used to begin the exploration. First, all neighboring nodes of the starting node are visited, then the neighboring nodes of those neighboring nodes are visited, and so on. During the exploration, each node and edge visited, as well as their degree of association with the starting node, are recorded. The degree of association can be measured in various ways, such as edge weights and the similarity between nodes. This information is very helpful for subsequent analysis of the importance of associated paths.

[0180] Simultaneously, during the exploration process, it is necessary to screen for association paths that meet the preset mining value based on established evaluation criteria. These evaluation criteria may include node attributes (such as node importance, activity, etc.), edge types (such as edges representing causal relationships, edges representing similarity relationships, etc.), or edge weights. By setting these criteria, it can be ensured that the explored association paths are meaningful, rather than valueless random paths. Optionally, the explored association paths can also be visualized.

[0181] After exploring all potential related paths, all candidate nodes directly or indirectly connected to the starting node are analyzed and recorded. These candidate nodes represent other anomalous information associated with the anomalous information. The associated anomalous information is then sorted and filtered based on the degree of association and the path length of the potential related paths, generating a filtered list of associated anomalous information.

[0182] S1043. Based on the correlation between the second abnormal entity corresponding to the abnormal information, the first abnormal entity, and the visualization data matrix of the first knowledge graph and the visualization data matrix of the second knowledge graph, determine at least one processing decision entity in the second knowledge graph that corresponds to the first abnormal entity and the second abnormal entity.

[0183] The server needs to determine the processing decision entity in the second knowledge graph based on the second abnormal entity (e.g., "wing crack"), the first abnormal entity ("wing structural fatigue"), and the correlation between the two knowledge graphs.

[0184] By comparing the visualized data matrices of the two graphs, the server discovered a strong correlation between "wing crack repair procedure" and both "wing crack" and "wing structural fatigue." Furthermore, "structural fatigue detection and prevention" was also closely related to these two anomalous entities. Therefore, the server identified these two processing decision entities as crucial for subsequent processing.

[0185] S1044. Determine the processing decision corresponding to the abnormal information based on the processing decision corresponding to the processing decision entity.

[0186] The server determines the specific processing decision for wing crack anomalies based on the identified processing decision entities (such as "wing crack repair program" and "structural fatigue detection and prevention").

[0187] For example, for the "wing crack repair procedure," the server can retrieve detailed step-by-step instructions, a list of required tools and materials, and safety operating guidelines. This information will directly guide maintenance personnel in performing the crack repair work.

[0188] Optionally, the “structural fatigue detection and prevention” treatment decision can also provide a systematic set of detection methods and preventive measures, aimed at preventing similar crack anomalies from recurring in the future.

[0189] The technical solution described in this embodiment combines knowledge graph and natural language processing technologies to provide intelligent processing and decision support for abnormal information discovered during aircraft inspection, thereby improving the efficiency and safety of aircraft maintenance.

[0190] In one possible implementation, determining the processing decision corresponding to the abnormal information based on the processing decision corresponding to the processing decision entity may further include:

[0191] The system acquires the maintenance history, spare parts inventory, and maintenance support information of the aircraft to be inspected. The maintenance history includes, but is not limited to, past maintenance records, parts replacement details, and results of periodic inspections. Analyzing this information helps identify which parts of the aircraft frequently experience problems and which components have reached or are nearing the end of their service life, thus aiding in predicting potential future issues. Spare parts inventory is crucial for understanding the types, quantities, and storage status of currently available spare parts. If a decision requires replacing a specific part but it is not in stock, the decision cannot be implemented immediately. Maintenance support information can include the number and skill levels of available maintenance personnel, as well as available maintenance equipment and tools. If a decision requires specific skills or equipment that are currently unavailable, the decision also cannot be implemented.

[0192] Based on the maintenance history information, spare parts inventory status, repair support information, and the processing decision corresponding to the processing decision entity, the processing decision is adjusted. The collected maintenance history information, spare parts inventory status, and repair support information are compared and analyzed with the preliminary processing decision corresponding to the processing decision entity to assess the feasibility and efficiency of the preliminary processing decision, and necessary adjustments are made to the decision considering the actual situation. Based on the above analysis, if it is found that the preliminary processing decision does not match the current situation, adjustments are necessary. For example, if the part to be replaced in the preliminary decision is not currently in stock, it may be necessary to use other available spare parts, or adjust the maintenance plan to wait for the spare parts to arrive. Similarly, if the maintenance personnel or equipment required in the preliminary decision are currently unavailable, it may be necessary to adjust the work plan or seek external support.

[0193] The adjusted processing decision is then determined as the processing decision corresponding to the abnormal information. After the above adjustments, the resulting processing decision will better reflect the current situation, and can then be determined as the final processing decision for the abnormal information. This decision not only considers the condition of the aircraft itself, but also comprehensively takes into account practical operational factors such as maintenance resources and spare parts inventory, thus making it more feasible and efficient.

[0194] The method provided in this application embodiment can ensure that the processing decisions made for each abnormal information are based on comprehensive information analysis and actual situation considerations, thereby improving the accuracy and efficiency of maintenance work.

[0195] Optionally, the method may also include content from multi-terminal interactions to provide the ability to handle emergency anomalies manually in real time or to handle anomalies that exist in automated processes.

[0196] Specifically, if there is abnormal feedback information regarding the processing decision corresponding to the abnormal information, and a call request is received from the inspection terminal, then based on the call request and the status of the maintenance terminal, an idle maintenance terminal is called. The call request is used to request a call with the maintenance terminal to provide abnormal feedback. If the idle maintenance terminal accepts the call request, a call connection is established between the inspection terminal and the idle maintenance terminal. If the idle maintenance terminal does not respond to the call request, other idle maintenance terminals are called until a call connection is established between the inspection terminal and the other idle maintenance terminals.

[0197] This application also provides an aircraft inspection system. The aircraft inspection system may include an inspection terminal, a server, and a maintenance terminal. The server can implement the method steps described in the foregoing method embodiments. The inspection terminal is used by inspection personnel to inspect the aircraft to be inspected and collect inspection images. The inspection terminal uploads the inspection images and information collected during the inspection process to the server. The server generates an inspection report based on the inspection images and information and sends the inspection report to the maintenance terminal. The maintenance terminal is used by back-end maintenance personnel to monitor and assist the aircraft inspection process and decision-making execution process based on the inspection report.

[0198] In one possible implementation, the server may include a proxy application for the real-time mobile inspection service, a core routing service, a real-time mobile inspection main service, a backend application for the inspection management platform, a file transfer service, a recording service, etc.

[0199] The proxy application for the mobile inspection real-time service is a middleware layer employing a load-balanced architecture to ensure stable and reliable service under high concurrency. Its main task is to receive connection requests from inspection terminals and forward them to the main mobile inspection real-time service on maintenance terminals. This proxy application allows inspection terminals to establish persistent connections with the server via the WebSocket protocol, enabling bidirectional transmission of real-time data.

[0200] The core routing service is the central nervous system of the entire system, responsible for managing user login / logout, state transitions, client routing requests, client connection management, and the management of LiveKit rooms (rooms for audio and video calls) and events (receiving and processing webhook events from LiveKit, such as changes in room state). It is also responsible for receiving and applying the system's dynamic configuration from Redis and publishing various state events to Redis for use by other applications or services.

[0201] The mobile inspection real-time main service is the core service for processing real-time inspection data. It provides a WebSocket interface to support real-time data exchange, including the exchange of audio and video streams, text messages, and other data. Inspection terminals can connect to the mobile inspection real-time service through a proxy application.

[0202] The backend application of the inspection management platform is the core system for managing inspection tasks and data. It adopts a load-balanced architecture to ensure high performance and scalability, while providing HTTP interfaces for frontend or other services to call. For example, it can provide interfaces for managing inspection tasks, querying inspection data, configuring inspection plans, etc.

[0203] The file transfer service handles file upload and download requests on the server, particularly for images (such as inspection photos) and documents (such as inspection-related documents reported by inspection personnel). It provides image upload functionality and returns the image's Uniform Resource Locator (URL) for later use. It also supports file download services.

[0204] The recording service is responsible for recording the audio and video content during the inspection process, offering two recording modes: room-wide recording and track-specific recording. Room-wide recording combines the video and audio of all participants into a single audio and video file. Track-specific recording, on the other hand, records the audio and video streams of each participant separately.

[0205] Optionally, the aforementioned aircraft inspection system can also support multi-tenant management. That is, the aircraft inspection system adopts a multi-tenant architecture, with tenants including the database, inspection terminals, inspection images captured by the terminals, video recorded by the server, and the server itself. Inspection database instances can be named according to tenant identifiers; different tenants can use different types of database instances, or they can use the same database instance.

[0206] Tenant segmentation for inspection terminals can be achieved by setting tenant numbers, which can be configured or changed through the terminal's initial page or settings interface. When uploading and downloading inspection images from the terminal, the messages used can carry a token obtained after login. The server uses the token to retrieve tenant information and creates a tenant directory in the root directory, achieving isolation between tenants. Similarly, the server can use a similar mechanism to achieve tenant isolation when recording audio and video of room conversations.

[0207] One possible implementation is that the server can distinguish tenants through the URL, which may carry a tenant mask. This tenant mask can be either a tenant identifier or a custom masquerading identifier.

[0208] Figure 2 This is a schematic diagram of an electronic device provided in an embodiment of this application. The electronic device is used to perform the aforementioned aircraft inspection method, and may be, for example, the aforementioned server. Figure 2 As shown, the electronic device 200 may include at least one processor 201, a memory 202, and a communication interface 203.

[0209] The memory 202 is used to store programs. Specifically, the program may include program code, which includes computer operation instructions.

[0210] The memory 202 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0211] The processor 201 is used to execute computer execution instructions stored in the memory 202 to implement the method described in the foregoing method embodiments. The processor 201 may be a CPU, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0212] The processor 201 can communicate and interact with external devices through the communication interface 203. These external devices can be, for example, the aforementioned inspection terminal or maintenance terminal. In specific implementations, if the communication interface 203, memory 202, and processor 201 are implemented independently, they can be interconnected via a bus to complete communication. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.

[0213] Optionally, in a specific implementation, if the communication interface 203, memory 202 and processor 201 are integrated on a single chip, then the communication interface 203, memory 202 and processor 201 can communicate through an internal interface.

[0214] This application also provides a computer-readable storage medium, which may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk. Specifically, the computer-readable storage medium stores program instructions, which are used in the methods described in the above embodiments.

[0215] This application also provides a program product including executable instructions stored in a readable storage medium. At least one processor of a computing device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the computing device to perform the actions of the above-described aircraft inspection method.

[0216] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. An aircraft inspection method, characterized in that, The method includes: The inspection requirements of each aircraft to be inspected are dynamically adjusted based on the inspection auxiliary information of each aircraft to be inspected. The inspection auxiliary information includes historical flight data, historical inspection data, and flight environment data. Based on the inspection requirements of each of the aircraft to be inspected, obtain inspection images of each of the aircraft to be inspected during the aircraft inspection process. The inspection images of each aircraft to be inspected are input into the inspection anomaly detection model to generate an inspection report for each aircraft to be inspected. The inspection report is used to indicate whether there are any anomalies on the surface of each aircraft to be inspected. If the inspection report includes abnormal information, then the first knowledge graph corresponding to the abnormal information and the second knowledge graph corresponding to the processing decision are obtained. The abnormal information is used to indicate that there is an abnormality on the surface of the aircraft to be inspected. Based on the association path of the abnormal information in the first knowledge graph, the associated abnormal information is determined, and the associated abnormal information is the abnormal information corresponding to the first abnormal entity that has an abnormal association with the abnormal information. Based on the correlation between the second abnormal entity corresponding to the abnormal information, the first abnormal entity, and the visualization data matrix of the first knowledge graph and the visualization data matrix of the second knowledge graph, at least one processing decision entity in the second knowledge graph corresponding to the first abnormal entity and the second abnormal entity is determined. Based on the processing decision corresponding to the processing decision entity, determine the processing decision corresponding to the abnormal information; The steps of obtaining the first knowledge graph corresponding to the abnormal information and the second knowledge graph corresponding to the processing decision include: An aviation knowledge base is pre-established, which contains knowledge data on aircraft components, common faults, maintenance procedures, and safety standards. The aviation knowledge base is generated by organizing and compiling technical documents from aircraft manufacturers, professional books in the aviation field, historical inspection records, and maintenance reports. Using natural language processing algorithms, entity recognition and relation extraction are performed on the abnormal information to determine the key entity nouns in the abnormal information and the relationships between the key entity nouns, and a first processing result is generated. Based on the first processing result, a first knowledge graph is constructed. In the first knowledge graph, each node represents a key entity noun, and the edges represent the relationships between the key entity nouns. In addition, decision data related to the processing decision is acquired, and the decision data is derived from maintenance manuals, best practice guidelines, and historical processing cases; The decision data is subjected to entity recognition and relation extraction using natural language processing algorithms to generate a second processing result. Based on the second processing result, a second knowledge graph is constructed. In the second knowledge graph, each node represents a processing strategy, maintenance tool or spare part, and the edge represents the logical relationship or dependency relationship between nodes. After generating the first knowledge graph and the second knowledge graph, the association between the first knowledge graph and the second knowledge graph is established, specifically by adding cross-knowledge graph edges to the first knowledge graph and the second knowledge graph; The step of determining the associated abnormal information based on the association path of the abnormal information in the first knowledge graph includes: By matching keywords in the abnormal information, the target node corresponding to the abnormal information is located in the first knowledge graph. The target node is used as the starting point, and potential related paths are explored along the edges starting from the starting point node. In the first knowledge graph, nodes are connected by the edges, which represent the relationship between nodes. During the exploration process, each node and edge visited is recorded, as well as the degree of association between each node and edge and the starting node. The degree of association is represented by weights. After exploring all potential related paths, analyze and record all candidate nodes that are directly or indirectly connected to the starting node. The candidate nodes represent other abnormal information associated with the abnormal information. The association anomaly information is sorted and filtered according to the degree of association and the path length of the potential association path to generate filtered association anomaly information; The step of locating the target node corresponding to the abnormal information in the first knowledge graph by matching keywords in the abnormal information includes: Keywords are extracted from the abnormal information, and the keywords include at least the aircraft component name, the fault type, and the abnormal phenomenon. After extracting the keywords, the keywords are preprocessed, including stop word removal and stemming. A word embedding algorithm is used to map the preprocessed keywords into a high-dimensional vector space to generate keyword vectors; The first knowledge graph is traversed using an approximate nearest neighbor search algorithm, and the cosine similarity between each node and the keyword vector is calculated. If the cosine similarity between any node and the keyword vector is greater than a set threshold, the node is determined as the target node corresponding to the abnormal information. Furthermore, if multiple nodes have a cosine similarity greater than a set threshold, then the nodes are sorted according to the magnitude of the cosine similarity, and the node with the highest similarity is selected as the target node. The step of taking the target node as the starting point and exploring potential associated paths along the edges from the starting node includes: Initialize the exploration environment, which specifically includes loading the complete graph data of the first knowledge graph and setting the corresponding exploration parameters, including the maximum exploration depth and the step size of each exploration. A breadth-first search algorithm is used to start from the target node and explore along the edges connected to it. Every node and edge visited is recorded, as well as the degree of association between each node and edge and the starting node. During the exploration process, associated paths that meet the preset mining value are selected based on the set evaluation criteria. The evaluation criteria include evaluation criteria based on node attributes, edge type or weight. The discovered connections will be visualized.

2. The method according to claim 1, characterized in that, The step of dynamically adjusting the inspection requirements of each aircraft to be inspected based on the inspection auxiliary information of each aircraft to be inspected includes: The inspection auxiliary information is obtained, and an inspection demand analysis network is established, wherein the inspection demand analysis network is used to determine the inspection demand of each of the aircraft to be inspected based on the inspection auxiliary information. Anomaly trend analysis is performed on the inspection auxiliary information based on the inspection demand analysis network. Based on the historical flight data and / or historical inspection data and / or flight environment data, a set of high-risk monitoring points for the aircraft to be inspected is determined, as well as the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points. Based on the set of high-risk monitoring points and the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points, the inspection requirements of each of the aircraft to be inspected are adjusted.

3. The method according to claim 2, characterized in that, The step of performing anomaly trend analysis on the inspection auxiliary information based on the inspection demand analysis network, determining the set of high-risk monitoring points for the aircraft to be inspected based on historical flight data, and / or historical inspection data, and / or flight environment data, and the inspection priority of high-risk monitoring points in the set of high-risk monitoring points, includes: The inspection demand analysis network determines the anomaly score calculation data for each aircraft component feature node based on the inspection auxiliary information and the mapping relationship between the inspection auxiliary information and the aircraft component feature nodes. The correlation analysis network is used to analyze the correlation between the anomaly score calculation data and the abnormal state of the aircraft component to obtain the degree of deviation between the current state and the normal state of the characteristic node of the aircraft component, as well as the abnormal trend. The deviation between the current state and the normal state of the aircraft component feature node, as well as the abnormal trend, are fused and calculated to generate an abnormal score for the aircraft component feature node. Based on the abnormal scores of the aircraft component feature nodes and the abnormal score threshold, the aircraft component feature nodes that are higher than the abnormal score threshold are taken as the set of high-risk monitoring points of the aircraft to be inspected. Based on the anomaly scores of each high-risk monitoring point in the set of high-risk monitoring points, the inspection priority of the high-risk monitoring points in the set of high-risk monitoring points is determined.

4. The method according to claim 3, characterized in that, The step of acquiring inspection images of each of the aircraft to be inspected during the inspection process, based on the inspection requirements of each of the aircraft to be inspected, includes: Based on the inspection requirements of each of the aircraft to be inspected, the inspection image parameters to be collected for different aircraft component feature nodes of each of the aircraft to be inspected are determined. The inspection image parameters include the number of inspection images, and / or the angle of the inspection images, and / or the clarity of the inspection images. The intelligent image capture model extracts the acquisition features of the inspection image parameters, identifies and classifies the image features, and generates image acquisition parameters for acquiring inspection images of each of the aircraft to be inspected. The inspection images of each of the aircraft to be inspected are acquired according to the image acquisition parameters.

5. The method according to claim 4, characterized in that, The anomaly detection model includes a visual recognition model and a deep learning model. The step of inputting the inspection images of each aircraft to be inspected into the anomaly detection model to generate an inspection report for each aircraft includes: The visual recognition model is used to extract image features from the inspection images. The image features include the shape, texture, and color of the aircraft parts of the aircraft to be inspected. The image features in the inspection image are mapped to the high-dimensional space of the deep learning model through a trained deep learning model to generate the feature vector corresponding to the image features. Based on the similarity calculation results between the feature vector corresponding to the image feature and the standard feature vector, the abnormal features of each aircraft component feature node of the aircraft to be inspected are determined, and the abnormal state and / or abnormal position corresponding to the abnormal features are marked. Based on the abnormal characteristics of the characteristic nodes of each aircraft component of each aircraft to be inspected, an inspection report is generated for each aircraft to be inspected.

6. The method according to claim 1, characterized in that, The step of determining the processing decision corresponding to the abnormal information based on the processing decision corresponding to the processing decision entity includes: Obtain the maintenance history, spare parts inventory, and maintenance support information of the aircraft to be inspected; The processing decision is adjusted based on the maintenance history information, the spare parts inventory, the repair support information, and the processing decision corresponding to the processing decision entity. The adjusted processing decision is determined as the processing decision corresponding to the abnormal information.

7. The method according to any one of claims 1-6, characterized in that, The method further includes: If there is abnormal feedback information for the processing decision corresponding to the abnormal information, and a call request is received from the inspection terminal, then according to the call request and the status of the maintenance terminal, the idle maintenance terminal is called. The call request is used to request to communicate with the maintenance terminal to provide abnormal feedback. If the idle maintenance terminal accepts the call request, a call connection is established between the inspection terminal and the idle maintenance terminal; If the idle maintenance terminal does not respond to the call request, it will call other idle maintenance terminals until a call connection is established between the inspection terminal and the other idle maintenance terminals.

8. An electronic device, characterized in that, include: The processor includes a communication interface and a memory, wherein the processor is communicatively connected to the communication interface and the memory, respectively. The memory stores computer-executed instructions; The communication interface communicates and interacts with external devices. The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the aircraft inspection method as described in any one of claims 1 to 7.

Images