Civil aviation damage detection method, device and equipment and readable storage medium

Abstract

This application relates to a method, apparatus, device, and readable storage medium for civil aviation damage detection. The method includes: constructing a multi-layered data storage model that integrates aircraft damage data and maintenance business data; the multi-layered data storage model includes at least a feature extraction layer, a graph relationship layer, and a business association layer; obtaining the damage feature vector of the current damage to the target aircraft; if the damage feature vector does not meet preset damage conditions, querying the relationship graph of the graph relationship layer based on the target aircraft and the damage feature vector to determine the corresponding similar damage identifier, and obtaining the corresponding historical damage dataset based on the similar damage identifier; performing damage detection on the target aircraft based on the historical damage dataset to determine the damage detection result of the target aircraft. This method can improve the efficiency of civil aviation damage inspection and maintenance.

Description

Technical Field

[0001] This application relates to the field of civil aviation data processing technology, and in particular to a civil aviation damage detection method, apparatus, equipment, and readable storage medium. Background Technology

[0002] During aircraft operation, various systems and components may experience malfunctions or potential problems. Managing maintenance data allows for timely monitoring of the aircraft's technical status, identifying wear, aging, and potential malfunctions in equipment. For example, maintenance-related data such as engine vibration and temperature data reflects the engine's health condition, enabling early detection and intervention of anomalies to prevent escalation and flight accidents, thus ensuring the safety of passengers and crew. Effective maintenance data management provides maintenance personnel with accurate and comprehensive information, helping them better understand system operating principles and failure modes, thereby enabling efficient and accurate maintenance work.

[0003] To ensure civil aviation safety, damage detection is necessary. However, current methods for damage detection rely on manual measurement, which leads to low efficiency. Therefore, a solution that improves damage detection efficiency is needed. Summary of the Invention

[0004] Therefore, it is necessary to provide a civil aviation damage detection method, device, computer equipment, computer-readable storage medium, and computer program product that can improve the efficiency of civil aviation damage detection in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a method for detecting damage in civil aviation aircraft, including:

[0006] A multi-layered data storage model integrating aircraft damage data and maintenance business data is constructed; the multi-layered data storage model includes at least a feature extraction layer, a graph relationship layer, and a business association layer.

[0007] In response to a damage detection command for a target aircraft, the damage feature vector of the target aircraft is obtained from the feature extraction layer according to the damage identifier carried by the damage detection command.

[0008] If the damage feature vector does not meet the preset damage conditions, then the relationship graph of the graph relationship layer is queried according to the target aircraft and the damage feature vector to determine the corresponding similar damage identifier, and the corresponding historical damage dataset is obtained according to the similar damage identifier.

[0009] Based on the historical damage dataset, damage confidence enhancement detection is performed on the target aircraft to determine the damage detection result of the target aircraft.

[0010] In one embodiment, the damage feature vector includes at least geometric features, curvature features, and texture features. If the damage feature vector does not meet a preset damage condition, a matching historical damage dataset is queried from the relation graph of the relation layer based on the target aircraft and the damage feature vector, including:

[0011] If the geometric feature does not satisfy the preset geometric feature condition but the curvature feature satisfies the preset curvature condition, then it is determined that the damage feature vector does not satisfy the preset damage condition.

[0012] The first historical damage node that matches the target aircraft in the relation graph of the relation layer, and the second historical damage node that matches the damage feature vector are determined.

[0013] Based on the first historical damage node and the second historical damage node, the corresponding damage data is obtained from the multi-layer data storage model to determine the historical damage dataset; the damage data corresponding to each historical damage node includes instance information of a complete historical damage.

[0014] In one embodiment, the step of determining the first historical damage node matched with the target aircraft from the relation graph of the relation layer, and the second historical damage node matched with the damage feature vector, includes:

[0015] Determine the damage location of the damage feature vector on the target aircraft, and the aircraft identifier of the target aircraft;

[0016] Based on the damage location and / or the aircraft identifier, determine the first matching historical damage node from the relation graph of the relation layer;

[0017] Calculate the similarity between the historical damage feature vector corresponding to the historical damage node in the relationship graph and the damage feature vector, and determine the historical damage node whose similarity is within a preset similarity range as the second historical damage node.

[0018] In one embodiment, the step of performing damage confidence enhancement detection on the target aircraft based on the historical damage dataset to determine the damage detection result of the target aircraft includes:

[0019] For each historical damage in the historical damage dataset, determine the weighted similarity between the historical damage feature vector and the damage feature vector of each historical damage.

[0020] Based on the weighted similarity, candidate historical damage datasets are determined from the historical damage dataset;

[0021] For each candidate historical damage in the candidate historical damage dataset, time series evolution analysis is performed based on the historical scan record data of each candidate historical damage to determine a second damage evolution pattern that matches the damage feature vector.

[0022] Based on the second damage evolution mode, the damage feature vector is predicted to obtain damage trend prediction results and prediction confidence data;

[0023] Based on the damage trend prediction results and the prediction confidence data, the damage detection results of the target aircraft are determined.

[0024] In one embodiment, the step of performing time-series evolution analysis on each candidate historical damage in the candidate historical damage dataset, based on the historical scan record data of each candidate historical damage, to determine a second damage evolution pattern matching the damage feature vector, includes:

[0025] For each candidate historical damage in the candidate historical damage dataset, historical scan record data for each candidate historical damage is extracted from the multi-layer data storage model; the historical scan record data is time series data.

[0026] For each of the time series data, the time series data is segmented, and the expansion rate of each time series data segment is calculated to obtain an expansion rate sequence;

[0027] Based on the expansion rate sequence and the time series data segment, identify the points of change in the expansion pattern;

[0028] Based on the expansion pattern change points and the expansion rate sequence, a first damage evolution pattern is determined for each candidate historical damage, and a second damage evolution pattern matching the damage feature vector is determined from the determined first damage evolution patterns.

[0029] In one embodiment, identifying the points of change in the expansion pattern based on the expansion rate sequence and the time series data segment includes:

[0030] For each time point in the time series data segment, the damage prediction error corresponding to each time point is determined sequentially. By performing cumulative deviation processing on the damage prediction error, candidate expansion mode change points are determined.

[0031] The candidate expansion mode change point is determined based on the expansion rate sequence, and the expansion rate ratio before and after the change is calculated.

[0032] The candidate expansion mode change points are verified and the expansion modes are classified based on the expansion rate ratio to determine the expansion mode change points.

[0033] In one embodiment, determining a first damage evolution pattern for each candidate historical damage based on the expansion pattern change point and the expansion rate sequence, and determining a second damage evolution pattern matching the damage feature vector from the determined first damage evolution patterns, includes:

[0034] Based on the expansion rate sequence and the expansion mode change points, an evolution feature vector is constructed;

[0035] Clustering is performed on the evolution feature vectors to determine the first damage evolution pattern corresponding to each candidate historical damage and the class center feature corresponding to each first damage evolution pattern.

[0036] Calculate the distance between the damage feature vector and each of the class center features, and determine a second damage evolution pattern that matches the damage feature vector from the first damage evolution pattern based on the distance.

[0037] In one embodiment, the step of predicting the damage feature vector based on the second damage evolution mode to obtain damage trend prediction results and prediction confidence data includes:

[0038] Obtain a target prediction model that matches the second damage evolution pattern;

[0039] The second damage evolution pattern and the damage feature vector are input into the target prediction model, and the predicted damage depth and prediction confidence interval of the target aircraft are output.

[0040] In one embodiment, the construction of a multi-layered data storage model that integrates aircraft damage data and maintenance operational data includes:

[0041] Acquire mesh data of the aircraft surface and 3D point cloud data for characterizing the details of the aircraft surface;

[0042] The three-dimensional point cloud data and the mesh data are analyzed to extract aircraft damage data of the damaged area from the three-dimensional point cloud data and to determine the damage identifier of the damaged area. The aircraft damage data includes damage geometric features and texture features.

[0043] Extract location information and scan metadata associated with the damaged area from the grid data;

[0044] The aircraft damage data and the scan metadata of the damaged area are stored in the feature extraction layer in a columnar storage structure, and the corresponding damage identifier is returned.

[0045] Multiple target data tables associated with aircraft maintenance business data are obtained. In the business association layer, the association relationship between the target data tables is established through foreign key relationships to associate the aircraft damage data and the aircraft maintenance business data; the damage table in the target data table is associated with the damage identifier.

[0046] Entity identification is performed on multiple target data tables. By determining the relationships between entities corresponding to the multiple target data tables, a relationship graph is constructed in the graph relationship layer to obtain a multi-layer data storage model.

[0047] In one embodiment, the step of performing entity recognition on multiple target data tables, and constructing a relationship graph in the graph relationship layer by determining the association relationships between entities corresponding to the multiple target data tables, includes:

[0048] Entity identification is performed on multiple target data tables to determine the entities corresponding to the multiple target data tables and the relationships between the entities; the entities include at least aircraft, damage, components, work cards, personnel, and manual chapters.

[0049] The entities are mapped to nodes in the graph, and edges for connecting their respective nodes are determined based on the association relationships. A relation graph is constructed in the graph relation layer based on the nodes and the edges.

[0050] Secondly, this application also provides a civil aviation damage detection device, comprising:

[0051] The model building module is used to construct a multi-layer data storage model that integrates aircraft damage data and maintenance business data; the multi-layer data storage model includes at least a feature extraction layer, a graph relationship layer, and a business association layer.

[0052] The data acquisition module is used to respond to a damage detection command for the target aircraft and, based on the damage identifier carried by the damage detection command, obtain the damage feature vector of the target aircraft from the feature extraction layer.

[0053] The graph query module is used to query the relationship graph of the graph relationship layer based on the target aircraft and the damage feature vector if the damage feature vector does not meet the preset damage conditions, determine the corresponding similar damage identifier, and obtain the corresponding historical damage dataset based on the similar damage identifier.

[0054] The damage detection module is used to perform damage confidence enhancement detection on the target aircraft based on the historical damage dataset, and to determine the damage detection result of the target aircraft.

[0055] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps described in any of the above method embodiments.

[0056] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps described in any of the above method embodiments.

[0057] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps described in any of the above method embodiments.

[0058] The aforementioned civil aviation damage detection methods, devices, computer equipment, computer-readable storage media, and computer program products construct a multi-layered data storage model that integrates aircraft damage data and maintenance business data. Based on this multi-layered data storage model, when performing damage detection on a target aircraft, the damage feature vector of the target aircraft is obtained from the feature extraction layer according to the damage identifier carried by the damage detection command. This eliminates the need for cross-system and cross-database data queries. The data storage model based on this storage format can quickly obtain the damage feature vector corresponding to the damage identifier. Furthermore, if the damage feature vector does not meet the preset damage conditions, a relational graph is introduced to further assist in aircraft damage detection. That is, the relational graph of the relational layer is queried based on the target aircraft and the damage feature vector to determine the corresponding similar damage identifier. The corresponding historical damage dataset is obtained based on the similar damage identifier. Damage confidence enhancement detection is performed on the target aircraft based on the historical damage dataset to determine the damage detection result of the target aircraft. The entire process replaces the discrete experience comparison after manual measurement, avoids the time-consuming and laborious problem of manual processing, and improves the efficiency of civil aviation damage detection. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a flowchart illustrating a civil aviation damage detection method in one embodiment;

[0061] Figure 2 This is a flowchart illustrating the construction method of a multi-layer data storage model in one embodiment;

[0062] Figure 3 This is a flowchart illustrating a method for determining a historical damage dataset in one embodiment;

[0063] Figure 4 This is a flowchart illustrating a civil aviation damage detection method in another embodiment;

[0064] Figure 5 This is a flowchart illustrating a data processing method in one embodiment;

[0065] Figure 6 This is a flowchart illustrating a method for querying civil aviation data in one embodiment;

[0066] Figure 7 Here is a system architecture diagram of a civil aviation damage detection method in one embodiment;

[0067] Figure 8 This is a schematic diagram of the data writing timing in one embodiment;

[0068] Figure 9 This is a flowchart illustrating a civil aviation damage detection method in yet another embodiment;

[0069] Figure 10 This is a structural block diagram of a civil aviation damage detection device in one embodiment;

[0070] Figure 11 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0072] In one exemplary embodiment, such as Figure 1 As shown, a civil aviation damage detection method is provided, which is applied to... Figure 1 The following steps, 102 to 108, are used as an example of computer equipment.

[0073] Step 102: Construct a multi-layer data storage model that integrates aircraft damage data and maintenance business data; the multi-layer data storage model shall include at least a feature extraction layer, a graph relationship layer, and a business association layer.

[0074] The multi-layered data storage model can be a hierarchical structured data organization framework that integrates aircraft damage data and maintenance business data. It consists of a raw data layer, a feature extraction layer, a graph relationship layer, and a business association layer. The raw data layer stores raw data related to aircraft 3D damage, such as raw point cloud files obtained from aircraft structural damage scans, complete 3D models, damage site photos, and scan parameters. The feature extraction layer extracts usable structured information from the raw data in the raw data layer, such as damage geometric features, location information associated with the damaged area, and scan metadata. The business association layer binds the damage features from the feature extraction layer to their corresponding maintenance business features using a unified index (e.g., aircraft ID, part ID, damage ID). The graph relationship layer organizes all data relationships into a knowledge graph, clarifying the network relationships between all entities. For example, if the entities include A3 aircraft type → wing underside → crack damage → XX maintenance process → XX aircraft material → historical similar damage cases → maintenance worker, by determining the relationships between entities, the graph can directly search for all historical maintenance solutions for the same type and location of damage on the same aircraft type, determining the optimal handling method.

[0075] It should be noted that the relationship graph based on the atlas layer can link historical damage evolution to find historical damage at the same location on the same aircraft, typical damage at the same location on the same aircraft model, and the historical evolution of damage with similar characteristics. It can also perform evolution analysis and prediction. Evolution analysis can include analyzing the temporal changes of damage parameters, calculating the expansion rate and trend, and identifying points of change in expansion patterns. Prediction can include predicting future trends based on historical evolution, providing final treatment results for similar damage, and assessing the possible development path of current damage. For example, in cases of abnormal damage expansion patterns, if a sudden 50% increase in damage depth is detected, analysis based on the relationship graph, querying the historical evolution of similar damage, obtaining the normal expansion rate, and inspecting the aircraft for scanning quality and environmental factors, would result in a warning of possible measurement error and a recommendation for rescanning.

[0076] Understandably, by constructing a multi-layered data storage model to link aircraft damage data with maintenance business data, the slow query efficiency of cross-file and database joins can be avoided.

[0077] In one exemplary embodiment, such as Figure 2 As shown, a method for constructing a multi-layer data storage model is provided, including the following steps:

[0078] Step 202: Obtain mesh data of the aircraft surface and 3D point cloud data for characterizing the details of the aircraft surface.

[0079] The 3D point cloud data can be a set of discrete points collected by a 3D scanning device, representing the spatial coordinates (x, y, z) and optional attributes (such as reflection intensity and color) of the aircraft surface. The 3D point cloud data can include aircraft damage data, which can include the length, width, height, normal vector, reflection intensity (different parts of the aircraft may use different materials, and the reflection intensity is used to identify material changes), area, and volume of pits on the aircraft surface.

[0080] The mesh data includes a mesh model representing the aircraft surface. This mesh model can be a topologically connected polygonal surface representation consisting of vertices, edges, and faces. It serves as a continuous geometric structure to approximate the aircraft's shape and can provide a semantically interpretable structural framework for the aircraft surface, supporting spatial context modeling of damaged areas and component attribution determination. The mesh model is also determined based on scanned point cloud data. It can be understood that this point cloud data is only used to determine the mesh model. The 3D point cloud data refers to the point cloud data acquired in a second scan, based on the point cloud data obtained in the first scan that determined the mesh model.

[0081] For example, a scan file collected by a 3D scanning device from an aircraft is acquired. This scan file includes a point cloud file, a mesh file (i.e., mesh data), and a texture file. The acquired scan file is stored in the raw data layer of a hierarchical storage model. The point cloud file includes 3D point cloud data characterizing the details of the aircraft surface, and may include damage feature data. The mesh file includes a mesh model and scan metadata, including the scan time, handheld device model, and operator. The mesh model includes the aircraft's position information, which is determined by the fuselage area and the 3D coordinate system. The texture file includes texture data characterizing the aircraft's texture.

[0082] Step 204: Analyze the 3D point cloud data and mesh data, extract aircraft damage data of the damaged area from the 3D point cloud data, and determine the damage marker of the damaged area.

[0083] The aircraft damage data includes damage geometry features, texture features, and curvature features.

[0084] For example, 3D point cloud data and mesh data are analyzed to identify damaged areas on the aircraft surface, and corresponding damage geometric features and damage identifiers are extracted from the 3D point cloud data. Identifying damaged areas on the aircraft surface can be based on geometric deviations, curvature abrupt changes, normal discontinuities, or multimodal fusion segmentation models between the 3D point cloud data and the mesh model to locate abnormal surface regions. Abnormal surface regions can include pits, cracks, bulges, scratches, etc., on the aircraft surface. Damage geometric features can include length, width, depth, area, volume, and curvature distribution. Damage identifiers can be assigned a unique, non-repeatable identifier to each identified damaged area.

[0085] Step 206: Extract location information and scan metadata associated with the damaged area from the grid data.

[0086] Step 208: Store the aircraft damage data and scan metadata of the damaged area in a columnar storage structure to the feature extraction layer and return the corresponding damage identifier.

[0087] Columnar storage structure can be a database storage format that organizes data by columns (fields) for physical storage. It can be used to support efficient local reading, compressed storage, and vectorized computation of high-dimensional sparse damage geometric features. For example, there is a data table where each row represents a damage and each column represents a feature, such as depth, length, width, location coordinates, scan time, scanner, and machine type.

[0088] Damage identifier, or damage ID, is the ID of each pit point on the aircraft's plane coordinate system. Since there may be more than one pit on the aircraft's plane, each pit has a unique ID in the plane coordinate system of each scan.

[0089] Step 210: Obtain multiple target data tables associated with aircraft maintenance business data; establish the relationship between the target data tables through foreign key relationships at the business association layer to associate aircraft damage data and aircraft maintenance business data; associate the damage table in the target data table with the damage identifier.

[0090] The system includes multiple target data tables, such as a damage table, an aircraft type table, a maintenance record table, and a manual chapter table. The damage table is associated with damage IDs from the feature extraction layer; the aircraft type table stores aircraft structural standards and tolerance data; the maintenance record table is associated with maintenance work cards, replaced parts, and personnel; and the manual chapter table is associated with relevant chapters of the airworthiness manual. The damage table can be a logical table in a relational database, using the damage identifier as the primary key and storing damage business attributes. The aircraft type table can be a static master data table recording the aircraft model, series, configuration code, applicable parts list, and service restrictions. The maintenance record table can be a transactional log table recording the execution time, personnel, work card number, replaced parts, inspection results, and release signature for each maintenance operation. The manual chapter table can be a structured reference table storing chapter numbers, titles, applicable conditions, inspection methods, and tolerance values ​​from maintenance manuals (such as SRM, AMM, IPC).

[0091] Foreign key relationships can be a logical association mechanism between tables implemented through field value reference constraints in relational databases. They ensure that the value of the referenced field exists in the primary key or unique key of the referenced table. They can be used to enforce the integrity of references between business data, and enable cross-table join queries to have semantic correctness and execution determinism.

[0092] For example, the damage table, aircraft type table, maintenance record table, and manual chapter table are obtained, and a many-to-many relationship is established between damage, aircraft type, record, and manual based on foreign key relationships to link three-dimensional damage data and aircraft maintenance business data.

[0093] Step 212: Entity identification is performed on multiple target data tables. By determining the relationships between the entities corresponding to the multiple target data tables, a relationship graph is constructed in the graph relationship layer based on the relationships to obtain a multi-layer data storage model.

[0094] Entities include at least aircraft, damage, components, work cards, personnel, and manual chapters. Entities serve as the basic semantic units of the relational graph, abstracting homogeneous information scattered across different target data tables into unified, referable objects. Entities are identified by parsing the table names, field names, field comments, and sample values ​​of the target data tables, combined with a predefined civil aviation ontology dictionary, to determine the domain entity type referred to by the field values. For example, entities include aircraft, damage, components, work cards, personnel, and manuals. Corresponding nodes can include aircraft nodes, damage nodes, component nodes, work card nodes, personnel nodes, and manual chapter nodes. The relationships between entities can be business semantic connections existing between different entity instances, including structural affiliation, process dependency, rule reference, and execution attribution, which can be used as the semantic basis for graph edge generation. Determining relationships can involve identifying the relationships between entities, analyzing foreign key constraints between tables, semantic co-occurrence of fields (e.g., a maintenance record table containing both damage_id and workcard_no), and maintenance procedure text, summarizing semantic relationship patterns such as "damage-trigger-work card" and "work card-reference-manual chapter." Relationships can include terms such as "belongs to," "cause," "reference," and "process." A relationship graph can be a directed attribute graph consisting of entities as nodes and semantic relationships as edges.

[0095] Based on the established relationship graph, when performing data queries, for example, executing a Cypher query: MATCH(p: damage) -[: located at] -> (c: part) <-[: belongs to] -(a: aircraft), (p) -[: trigger] -> (w: workcard) -[: reference] -> (m: manual chapter) WHERE c.part_no = 'FAN_BLADE_3RD_STAGE' AND m.chapter_id = 'AMM72-31-00' RETURNDISTINCT Ta.reg_num, w.workcard_no, m.title; a single query returns all damage events involving the part and manual chapter and their associated workcards, without the need for cross-database JOIN or manual comparison.

[0096] For example, entity recognition is performed on multiple target data tables to determine the entities corresponding to the multiple target data tables and the relationships between the entities; the entities are mapped to nodes of the graph, and the edges used to connect their respective nodes are determined according to the relationships. A relationship graph is constructed in the graph relationship layer according to the relationships. By constructing a four-layer storage model of "raw data layer - feature extraction layer - business relationship layer - graph relationship layer", a multi-layer data storage model is obtained.

[0097] Based on the relationship graph constructed above, when an incremental dataset is detected in the target data table, the incremental dataset is identified to determine the target entity corresponding to the incremental dataset and the relationship between the target entities; if there is no node corresponding to the target entity in the relationship graph, the node corresponding to the target entity and the edge used to represent the relationship between the target entities are created in the relationship graph.

[0098] Incremental datasets can be a set of records that are added to or modified based on an existing target data table. For example, detecting data changes in the business-related layer of a hierarchical storage model generates incremental data. Incremental datasets can be obtained, but are not limited to, by parsing database logs.

[0099] For example, by monitoring database transaction logs or application layer API call logs, identifying data write events that meet incremental conditions, obtaining the incremental dataset of the target data table, performing joint parsing on the field values ​​of the incremental dataset, comparing the identified entity types with existing entity ontologies, filtering out entities for which there are no corresponding instances in the graph, determining the target entities, and determining the relationships between target entities, and then creating corresponding nodes for the target entities and edges representing the relationships between the target entities in the relationship graph. For example, when a new entity or relationship is generated, the graph construction service converts the entity into a node in the graph database and the relationship into an edge; it uses the graph database's API to create or update nodes and edges. For instance, if the business layer creates a maintenance record recording that a certain damage is processed through a certain work card, nodes for this damage and this work card will be created in the relationship graph, and an edge will be created between them, with the edge type noted. In this approach, by detecting the existence of incremental datasets in the target data table, a lightweight awareness of dynamic changes in maintenance business is achieved. By identifying incremental datasets, the relationship graph can be updated in a timely manner, realizing the dynamic expansion of the relationship graph.

[0100] In the aforementioned embodiment of the multi-layer data storage model, data extraction is performed based on the raw data obtained from the aircraft surface mesh data and the 3D point cloud data used to characterize the aircraft surface details. The extracted damage feature data of the damaged area, along with associated location information and scan metadata, is stored in a columnar storage structure. This data is then associated with damage tables in multiple target data tables linked to aircraft maintenance business data through corresponding damage identifiers. Furthermore, foreign key relationships are used to establish connections between target data tables, and a related relationship graph is built. This achieves the fusion of 3D damage data and aircraft maintenance business data, addressing the lack of structured relationships between data in traditional storage methods. This storage approach shortens the response time of complex queries using multi-layered index structures, improving query efficiency and thus enhancing damage detection efficiency. It also avoids slow responses to cross-file and database joint queries, which cannot meet real-time interaction requirements. In addition, damage detection based on the relationship graph enables previously fragmented data fragments to achieve cross-entity relationships through the graph relationship layer, enhancing data correlation and improving the accuracy of damage detection.

[0101] Step 104: Obtain the damage feature vector of the target aircraft's current damage.

[0102] The damage feature vector can include a damage geometric feature set, a curvature feature set, and a texture feature set. Damage is identified by a corresponding damage area on the target aircraft, also known as the damage location. The damage feature vector of the current damage to the target aircraft can be determined by preprocessing the acquired 3D point cloud data, constructing a mesh model, and then performing feature extraction and damage differentiation. No specific limitations are placed on the method for determining the damage feature vector of the current damage.

[0103] Curvature in civil aviation damage detection includes Gaussian curvature, mean curvature, and principal curvature. The purpose of determining curvature is to extract multi-dimensional geometric features from the mesh model surface, providing a quantitative basis for automatic segmentation and type determination of damaged areas. Gaussian curvature describes the inherent bending characteristics of a surface, distinguishing basic surface types (elliptic points, hyperboloid points, parabolic points). In damage detection, pitted areas typically exhibit negative Gaussian curvature, while normal surface areas have Gaussian curvature close to zero or positive. This characteristic helps the system initially identify potential pitted areas. Principal curvature characterizes the direction and magnitude of maximum and minimum bending. In damage detection, the ratio and directionality of principal curvature can be used to distinguish between isotropic damage (such as pits) and anisotropic damage (such as scratches). Mean curvature describes the external bending degree of a surface, representing the average bending degree of the surface in the normal direction. In damage detection, the absolute value of the mean curvature reflects the depth of the damage, used for subsequent severity grading. Curvature calculation methods can be implemented using existing methods and will not be elaborated upon here.

[0104] Step 106: If the damage feature vector does not meet the preset damage conditions, query the relationship graph of the graph relationship layer based on the target aircraft and the damage feature vector to determine the corresponding similar damage identifier, and obtain the corresponding historical damage dataset based on the similar damage identifier.

[0105] In this case, if the damage feature vector does not meet the preset damage conditions and the confidence level of the damage feature vector is less than the preset confidence level value, the damage type of the current damage cannot be accurately determined. Damage types include pits, corrosion, etc. In this embodiment, pits are used as an example for illustration. The corresponding preset damage conditions are preset pit conditions, which include damage geometric feature conditions, curvature feature conditions, and texture feature conditions. Damage geometric feature conditions include: the depth-to-diameter ratio is less than a preset value (e.g., 0.5), the boundary is closed and approximately circular or elliptical, and the surface depression is continuous without breakage. Curvature feature conditions include negative curvature in the central region, positive curvature at the edges, and the curvature gradient gradually increasing from the edge to the center. Texture feature conditions include good surface texture continuity and no material loss or discoloration.

[0106] If the damage geometry condition in the preset damage conditions is met, it is initially determined to be a pit; if both the damage geometry condition and the curvature condition are met, it is confirmed to be a pit; if both the damage geometry condition and the texture condition are confirmed to be met, it is determined to be a minor pit. It is understood that the degree of minorness in this example is preset and will be determined according to actual needs, which will not be elaborated here.

[0107] If the damage geometry condition is not met but the curvature condition is met, it cannot be determined whether it is a true pit. This indicates that the preset damage condition is not met; that is, the curvature characteristic indicates abnormal bending on the surface, but the damage geometry is insufficient to definitively identify it as a pit, requiring further assessment. In this example, further assessment is performed by analyzing the relationship graph of the graph relationship layer in a multi-layer data storage model.

[0108] The methods for identifying similar damage identifiers can be as follows: querying the relation graph of the graph relation layer based on the target aircraft and the damage feature vector; or jointly encoding the target aircraft configuration information and the damage feature vector into a graph query vector, and performing subgraph matching or nearest neighbor retrieval in the relation graph to locate damage nodes with similar topological structures.

[0109] Similar damage identifiers can also be determined based on the relationship graph by identifying damage identifiers corresponding to the same aircraft type and damage area as the target aircraft, by calculating the similarity between the damage feature vector and the historical damage feature vector corresponding to the historical damage node in the relationship graph, or by retrieving the distance between the damage location corresponding to the historical damage node in the relationship graph and the damage location corresponding to the damage identifier.

[0110] The historical damage dataset includes instance information for at least one complete historical damage instance. This instance information includes at least complete features, the final assessment result, maintenance records, and the evolution process. For example, the acquired instance information may include damage ID, detection time, flight hours, damage type (e.g., dents, cracks, corrosion, delamination, etc.), damage size (length, depth, area), aircraft type, damage location, maintenance records, and the evolution process.

[0111] Step 108: Perform damage detection on the target aircraft based on the historical damage dataset and determine the damage detection results of the target aircraft.

[0112] The damage detection process can involve determining the weighted similarity between the historical damage feature vector and the damage feature vector for each historical damage in the historical damage dataset; determining candidate historical damage datasets based on the weighted similarity; and performing a voting statistics on the damage type corresponding to each candidate historical damage in the candidate historical damage datasets to determine the damage type of the current damage.

[0113] For example, for each historical damage in the historical damage dataset, a weighted similarity between the historical damage feature vector and the damage feature vector of each historical damage is determined; based on the weighted similarity, a candidate historical damage dataset is determined from the historical damage dataset; for each candidate historical damage in the candidate historical damage dataset, time series evolution analysis is performed on the historical scan record data of each candidate historical damage to determine a second damage evolution pattern that matches the damage feature vector; the damage feature vector is predicted based on the second damage evolution pattern to obtain damage trend prediction results and prediction confidence data; based on the damage trend prediction results and prediction confidence data, the damage type of the target aircraft is determined.

[0114] For example, if corrosion is detected as the current damage, but historical records show no corrosion at that location, analysis based on the relationship graph can query damage records for the same location on the same model and determine that the damage is a scratch. In this case, reverse verification is needed, re-detecting the characteristics of the current damage to confirm it as a scratch, and adjusting the damage detection results accordingly. Furthermore, by comparing damage patterns of the same model / location based on the relationship graph, cluster analysis of damage characteristics of the same model can identify high-frequency damage locations and types to determine systemic damage patterns. Then, comparing the current damage with damage of the same model can assess the typicality and particularity of the damage. This can identify design or manufacturing defects and provide maintenance references, such as standard handling procedures for damage of the same model, and successful maintenance experiences for similar damage.

[0115] The aforementioned civil aviation damage detection method constructs a multi-layered data storage model that integrates aircraft damage data and maintenance business data. Based on this model, when performing damage detection on a target aircraft, the damage feature vector of the target aircraft is obtained from the feature extraction layer according to the damage identifier carried by the damage detection command. This eliminates the need for cross-system and cross-database data queries. This data storage model can quickly obtain the damage feature vector corresponding to the damage identifier. Furthermore, if the damage feature vector does not meet the preset damage conditions, a relational graph is introduced for further auxiliary detection of aircraft damage. Specifically, the relational graph of the relational layer is queried based on the target aircraft and the damage feature vector to determine the corresponding similar damage identifier. The corresponding historical damage dataset is then obtained based on the similar damage identifier. Damage detection of the target aircraft is then performed based on the historical damage dataset to determine the damage detection result. This entire process replaces the discrete experience comparison after manual measurement, avoiding the time-consuming and labor-intensive problem of manual processing and improving the efficiency of civil aviation damage detection.

[0116] In one exemplary embodiment, a method for determining a historical damage dataset is provided, such as Figure 3 As shown, it includes the following steps:

[0117] Step 302: If the geometric features do not meet the preset geometric feature conditions but the curvature features meet the preset curvature conditions, then the damage feature vector is determined to not meet the preset damage conditions.

[0118] Step 304: Determine the first historical damage node that matches the target aircraft from the relation graph of the relation layer, and the second historical damage node that matches the damage feature vector.

[0119] The relationship graph includes historical damage nodes, each corresponding to an aircraft type and damage location. By determining the aircraft type of the target aircraft and the damage location corresponding to the damage identifier, the relationship graph can be used to query historical damage nodes that share the same aircraft type and / or damage location as the target aircraft, as well as historical damage nodes whose spatial proximity to the damage location corresponding to the damage identifier is within a preset distance range, thus obtaining the first historical damage node. By calculating the similarity between the damage feature vector and the historical damage feature vectors corresponding to each historical damage node in the relationship graph, historical damage nodes with similarity within a preset similarity range are determined as the second historical damage node.

[0120] Optionally, in an exemplary embodiment, determining a first historical damage node matching the target aircraft from the relation graph of the relation layer, and a second historical damage node matching the damage feature vector, includes:

[0121] The damage feature vector is determined at the damage location corresponding to the target aircraft, as well as the aircraft identifier of the target aircraft. Based on the damage location and / or aircraft identifier, a matching first historical damage node is determined from the relationship graph of the relationship layer. The similarity between the historical damage feature vector corresponding to the historical damage node in the relationship graph and the damage feature vector is calculated, and historical damage nodes with similarity within a preset similarity range are determined as second historical damage nodes. Each damage identifier has a corresponding damage feature vector and a corresponding damage location on its respective aircraft.

[0122] For example, historical damage nodes of the same model and location as the damage feature vector are retrieved from the graph relation layer, and other damage nodes that are spatially adjacent to the damage location of the damage feature vector (e.g., distance <50mm) are retrieved to obtain the first historical damage node; historical damage nodes with a similarity > threshold (e.g. 0.8) between the historical damage feature vector and the damage feature vector are retrieved from the graph relation layer to obtain the second historical damage node.

[0123] Step 306: Based on the first historical damage node and the second historical damage node, obtain the corresponding damage data from the multi-layer data storage model to determine the historical damage dataset; the damage data corresponding to each historical damage node includes instance information of a complete historical damage.

[0124] It is understandable that the multi-layer data storage model is a four-layer storage model built on the basis of "raw data layer - feature extraction layer - business association layer - graph relationship layer". Then, based on the historical damage identifiers corresponding to the first and second historical damage nodes, the corresponding damage data can be queried from the multi-layer data storage model to obtain the historical damage dataset.

[0125] In the above embodiments, when the damage feature vector does not meet the preset damage conditions, historical damage matching the damage feature vector is determined by searching the relationship graph. Based on the historical damage, the damage type of the current damage corresponding to the damage feature vector can be further accurately determined.

[0126] Understandably, based on the above, we can also calculate the feature similarity between each historical damage and the current damage corresponding to the damage feature vector based on the damage dataset to determine the candidate historical damage dataset; perform time series evolution analysis on each candidate historical damage data in the candidate historical damage dataset to determine the corresponding expansion pattern change points and expansion rate sequences; determine the damage evolution pattern that matches the damage feature vector based on the expansion pattern change points and expansion rate sequences; and make future trend predictions based on the matched damage evolution patterns and damage feature vectors to obtain the corresponding confidence intervals and future trend prediction results.

[0127] In one exemplary embodiment, such as Figure 4 As shown, a method for detecting damage in civil aviation aircraft is provided, including the following steps:

[0128] Step 402: For each historical damage in the historical damage dataset, determine the weighted similarity between the historical damage feature vector and the damage feature vector of each historical damage.

[0129] The historical damage feature vector and the damage feature vector both include damage geometric features, curvature features, and texture features. For each historical damage in the historical damage dataset, the similarity of feature vectors of the same dimension is calculated. The similarity corresponding to each dimension is weighted to obtain the weighted similarity between the historical damage feature vector and the damage feature vector for each historical damage. For example, for each historical damage hi, the weighted similarity S_i with the current damage is calculated: S_i = w1×sim(geometric features) + w2×sim(curvature features) + w3×sim(texture features); where w1, w2, and w3 are weight coefficients, which can be dynamically adjusted according to business rules and are not specifically limited here.

[0130] Step 404: Determine candidate historical damage datasets from the historical damage dataset based on weighted similarity.

[0131] For example, based on weighted similarity, historical damages with a weighted similarity greater than a preset similarity threshold are identified as candidate historical damages, and a candidate historical damage dataset is determined based on the candidate historical damage data corresponding to each candidate historical damage. Further, based on the damages corresponding to each candidate historical damage data in the candidate historical damage dataset, voting statistics are performed according to the determined damage types to obtain the statistical number of different damage types. Damage types include pits, scratches, corrosion, and normal damage, etc.

[0132] Step 406: For each candidate historical damage in the candidate historical damage dataset, perform time series evolution analysis based on the historical scan records of each candidate historical damage to determine the second damage evolution pattern that matches the damage feature vector.

[0133] The time series evolution analysis process involves extracting corresponding historical scan data from each candidate historical damage in the candidate historical damage dataset, calculating the expansion rate of the historical scan data to obtain an expansion rate vector (i.e., expansion rate sequence), and determining the expansion rate change points based on the expansion rate vector. The first damage evolution pattern is then determined based on the expansion rate change points and the expansion rate sequence. Finally, the damage evolution pattern matching the damage feature vector is determined based on the expansion pattern change points and the expansion rate sequence. This can be achieved by clustering the expansion pattern change points and expansion rate sequences corresponding to each candidate historical damage to determine the first damage evolution pattern, and then determining the second damage evolution pattern matching the damage feature vector from the first damage evolution pattern.

[0134] Furthermore, for each candidate historical damage in the candidate historical damage dataset, time series evolution analysis is performed based on the historical scan records of each candidate historical damage to determine a second damage evolution pattern that matches the damage feature vector. This includes: for each candidate historical damage in the candidate historical damage dataset, extracting the historical scan records of each candidate historical damage from a multi-layer data storage model; the historical scan records are time series data; for each time series data, segmenting the time series data, calculating the expansion rate of each time series data segment to obtain an expansion rate sequence; identifying expansion pattern change points based on the expansion rate sequence and the time series data segments; determining a first damage evolution pattern for each candidate historical damage based on the expansion pattern change points and the expansion rate sequence; and determining a second damage evolution pattern that matches the damage feature vector from the determined first damage evolution pattern.

[0135] The historical scan data includes parameters such as damage depth at different times; this example uses damage depth d as an example. The historical scan data is sorted by time to obtain the corresponding time series data. The time series data can be segmented according to a preset length; no specific limitation is made here.

[0136] The expansion rate of each time series data segment can be determined based on the number of time series data points (n) or on actual needs. Methods for determining the expansion rate include linear fitting, calculation using the difference method, and calculation using the sliding window method. For example, if the number of segments is greater than a preset number, the difference method is used. For instance, if the number of segments n < 3, the difference method is used to calculate the rate of each segment, k_i = (d_{i+1} - d_i) / (t_{i+1} - t_i) (i = 1ton-1), and the overall average rate is:

[0137] k_avg = (d_n - d_1) / (t_n - t_1). If the number of data points n ≥ 3, piecewise linear fitting can be used. That is, for the first 50% of data points, a straight line is fitted to obtain k_early; for the last 50% of data points, a straight line is fitted to obtain k_late; and for all data points, a straight line is fitted to obtain k_overall. Alternatively, a sliding window method can be used. By setting the window size W = min(3, floor(n / 2)), the slope of each window is calculated to form an expanding rate sequence. Further, based on the obtained expanding rate sequence, the maximum rate, minimum rate, rate method, and rate trend can be determined. For example, the maximum rate k_max = max(k_i), the minimum rate k_min = min(k_i), and the rate variance var_k = 1 / (n-1) × Σ(k_i - k_avg)². In an exemplary embodiment, if three consecutive segments show an increasing rate, the rate trend is marked as an "accelerating trend".

[0138] In an exemplary embodiment, identifying expansion pattern change points can be achieved by segmenting the time series data into several segments, calculating the slope of each segment, and determining the expansion pattern change point by comparing the relative slope changes of adjacent slopes. For example, the time series data is divided into several segments, and the slopes k1, k2, and ki of each segment are calculated. The relative slope changes of adjacent slopes are compared, i.e., the change point detection condition is: |k_{i+1}-k_i| / k_i>threshold T_change (e.g., 0.3). If the condition is met, the time point is marked as an expansion pattern change point. This time point can be, but is not limited to, the boundary point between two time series data segments.

[0139] In an exemplary embodiment, identifying points of change in the extended pattern can also be based on a sliding window. Within the window, a local slope k_window_i is fitted, and the slope changes of adjacent windows are compared to determine the points of change in the extended pattern. For example, firstly, the window size W=3 (at least 3 points are needed to calculate the trend); secondly, starting from the earliest data point, the window is slidable, and the slope k_window_i of the data within each window is calculated, along with the fitting error e_i; thirdly, slope change detection is performed, with the slope change rate Δk=|k_window_{i+1}-k_window_i| / max(|k_window_i|,ε); if Δk>th threshold T_change (e.g., 0.5), the (i+W / 2)th time point is marked. The potential change point is identified, and the slope before the change, k_before=k_window_i, and the slope after the change, k_after=k_window_{i+1}, are recorded. If the potential change point has more than a preset number of data points before and after the change (e.g., 2), and the change lasts for at least a preset number of time points (excluding single-point anomalies), the preset number of time points can be, but is not limited to, 2, and the fitting errors e_i and e_{i+1} are both less than the threshold T_error, then the potential change point is identified as an extended mode change point.

[0140] In an exemplary embodiment, identifying expansion pattern change points based on the expansion rate sequence and time series data may further include: for each time point in the time series data, sequentially determining the damage prediction error corresponding to each time point, and determining candidate expansion pattern change points by performing cumulative deviation processing on the damage prediction errors; determining the ratio of expansion rates before and after the expansion rate change points based on the expansion rate sequence; verifying and classifying the candidate expansion pattern change points based on the expansion rate ratio, and determining the expansion pattern change points.

[0141] Specifically, for each time point in the time series data, from the second time point to the last time point, the predicted damage depth for each time point is calculated based on the damage depth of the previous time point and the average spread rate of the time series data. The prediction error is determined based on the damage depth of the current time point and the corresponding predicted damage depth, and the prediction errors are accumulated. If the accumulated sum is greater than a threshold, the corresponding time point is marked as a candidate spread pattern change point, the accumulated sum is reset, and the change trend is recorded based on the prediction error. The number of data points before and after the change of the candidate spread pattern change point, as well as the ratio of the average spread rate before and after the change, are obtained. Based on the number of data points and the ratio of the average spread rate before and after the change, the candidate spread pattern change points are verified and classified according to the spread pattern, thus determining the spread pattern change point.

[0142] For example, initialize the cumulative sum S0 = 0. For each time point i (from 2 to n): Predict the damage depth d_pred = d_{i - 1}+k_avg×(t_i - t_{i - 1}); Predict the error e_i = d_i - d_pred; The cumulative sum S_i = max(0, S_{i - 1}+e_i - δ), where δ is the tolerance deviation (usually set to 1 / 2 of the noise level); If S_i > the threshold H (e.g., H = 5×noise level), mark the time point i as a candidate expansion mode change point, reset S_i = 0, and record the change type: If e_i > 0 and continuous: Expansion acceleration; If e_i < 0 and continuous: Expansion deceleration or repair. If there are at least 2 data points before and after the candidate expansion mode change point are detected, and the calculated ratio r = k_after / k_before of the average rates before and after the change is within the preset range, then determine the corresponding candidate expansion mode change point as an expansion mode change point. Further, if the ratio of the average rates before and after the change is within the first preset ratio range (e.g., r > 1.5), then determine the expansion mode as significant acceleration; if the ratio of the average rates before and after the change is within the second preset ratio range (e.g., 0.67 < r < 1.5), then determine the expansion mode as stable expansion; if the ratio of the average rates before and after the change is within the third preset ratio range (e.g., r < 0.67), then determine the expansion mode as significant deceleration; if the ratio of the average rates before and after the change is within the fourth preset ratio range (e.g., r is close to 0), then determine the expansion mode as tending to be stable. It can be understood that the determination degree of r being close to 0 is defined according to actual needs, which can be preset and will not be specifically described here.

[0143] The above-mentioned identification methods of expansion mode change points in different ways can match the requirements of different scenarios. By accurately identifying the expansion mode change points, the damage evolution mode matching the damage feature vector can be accurately identified, and thus the accuracy and reliability of civil aviation damage detection can be improved.

[0144] In an exemplary embodiment, a method for determining a second damage evolution mode matching a damage feature vector is provided:

[0145] According to the expansion mode change points and the expansion rate sequence, determine the first damage evolution mode of the candidate historical damage, and determine the second damage evolution mode matching the damage feature vector from the determined first damage evolution mode, including: Construct an evolution feature vector according to the expansion rate sequence and the expansion mode change points; Perform clustering processing on the evolution feature vector to determine the first damage evolution mode corresponding to the candidate historical damage and the class center features corresponding to each first damage evolution mode; Calculate the distances between the damage feature vector and each class center feature, and determine the second damage evolution mode matching the damage feature vector from the first damage evolution mode according to the distances.

[0146] The expansion rate sequence can be obtained as described above and will not be repeated here. The expansion rate sequence includes at least the average expansion rate, the maximum expansion rate, and the expansion rate corresponding to each time series data segment. The evolution feature vector includes at least the time from the first discovery to the first repair of each historical damage, the time from the first discovery to the threshold, the average expansion rate, the maximum expansion rate, the number of change points of the expansion pattern change points, whether acceleration or deceleration occurred, and the final disposal result (e.g., repair / monitoring / replacement).

[0147] The first damage evolution mode includes at least one or more of the following: first mode: slow stable type (k<0.01mm / thousand cycles, no change point), second mode: accelerated deterioration type (there is an accelerated change point, eventually requiring repair), third mode: early repair type (repaired quickly after initial discovery), and fourth mode: long-term stable type (no significant change after many years of monitoring).

[0148] For example, an evolution feature vector is constructed based on the expansion rate sequence and the expansion pattern change points; the evolution feature vector is clustered using the K-means clustering algorithm to determine the first damage evolution pattern and the class center features corresponding to each first damage evolution pattern; the distance between the damage feature vector and each class center feature is calculated, and the second damage evolution pattern matching the damage feature vector is determined from the first damage evolution patterns based on the distance.

[0149] Step 408: Predict the damage feature vector based on the second damage evolution mode to obtain damage trend prediction results and prediction confidence data.

[0150] For example, a target prediction model corresponding to the second damage evolution mode is determined, and initial parameters of the target prediction model are set. The damage depth of the current damage corresponding to the damage feature vector and the standard damage depth threshold of the target aircraft are input into the target prediction model, and the corresponding remaining lifetime and prediction interval are output. It is understood that different types of damage evolution modes have their own corresponding target prediction models. The number of target prediction models can be a single prediction model or a prediction model pool constructed by fusing multiple models. The prediction model pool can include linear extrapolation, exponential extrapolation, piecewise models, similar damage statistical models, etc. For example, linear extrapolation d(t)=d_current+k_avg×(t-t_current), where k_avg is the average expansion rate; exponential extrapolation d(t)=d_current×e^{λ×(t-t_current)}; linear extrapolation, exponential extrapolation, piecewise models, and similar damage statistical models can output their respective predicted values ​​for the same input t, that is, the output is the predicted damage depth at a future time t.

[0151] Optionally, a pool of prediction models (linear, exponential, piecewise, and similar damage statistics) built based on a multi-model fusion model is weighted according to their fit to the current damage and the second damage evolution pattern, and then their predicted damage depth is weighted and averaged; the confidence interval of the prediction is then calculated by simulating the random fluctuation of the parameters.

[0152] In an exemplary embodiment, the damage feature vector is predicted according to the second damage evolution mode to obtain damage trend prediction results and prediction confidence data, including: obtaining a target prediction model that matches the second damage evolution mode; inputting the second damage evolution mode and the damage feature vector into the target prediction model, and outputting the predicted damage depth of the target aircraft and the prediction confidence interval.

[0153] Specifically, a target prediction model matching the second damage evolution pattern is constructed, i.e., a prediction model pool is built. This pool includes four models: linear extrapolation, exponential extrapolation, a statistical model determined based on similar damage, and a pre-defined segmented model. Weights are assigned to the four models, with each weight determined based on its fitting error to historical data. The second damage evolution pattern and damage feature vector are used as inputs to the target prediction model to obtain the predicted damage depth of each model's output. The predictions from different models are summed according to their respective weights to obtain a fused prediction curve. Monte Carlo simulation is then used, allowing the model parameters to be repeatedly and randomly sampled within their uncertainty range for multiple predictions. The quantiles from the distribution formed by these prediction results are used to obtain the confidence interval for the future trend.

[0154] For example, construct a prediction model pool: Model 1: Linear extrapolation d(t) = d_current + k_avg × (t - t_current); Model 2: Exponential extrapolation d(t) = d_current × e^{λ × (t - t_current)}; Model 3: Statistical model based on similar damage; Model 4: Piecewise model (if there are change points); assign weights to each model and perform weight normalization: w_i = 1 / RMSE_i (based on the model's fitting error to historical data), normalize the weights... The formula is as follows: W_i = w_i / Σw_i; For a future time t, each model provides a predicted value d_i(t). The predicted values ​​of each model are fused to obtain the final weighted average d_pred(t): d_pred(t) = Σ[W_i × d_i(t)]; Further, for each model in the prediction model pool, considering the uncertainty of the parameters, N parameter sampling samples are generated. The predicted value is calculated for each sample to obtain the prediction distribution curve. The 2.5% and 97.5% quantiles are taken as the 90% confidence interval. It can be understood that by fusing multiple models for prediction, compared with a single model, the robustness of the model can be improved and different damage evolution modes and uncertainties can be taken into account.

[0155] In the above embodiments, by performing clustering processing on the constructed evolution feature vectors, the damage evolution patterns of candidate historical damages are classified, and by calculating the distance between the damage feature vectors and the central features of each type, the damage evolution pattern of the current damage can be accurately determined.

[0156] Furthermore, based on the damage trend prediction results and prediction confidence data, recommended handling strategies for the current damage can be generated, enabling timely repair, improving repair efficiency, and ensuring aircraft safety. Specifically, based on the candidate historical damage dataset, the final handling records for each candidate historical damage are determined. From each candidate historical damage, the top-K historical damages most similar to the current damage are identified, and the proportion of each handling method is statistically analyzed. Based on each proportion and its corresponding preset proportion, recommended handling strategies are generated. The final handling records can include maintenance records, continuous monitoring records, and replacement records. For example, maintenance records can include maintenance time, maintenance method, and post-maintenance effects. Continuous monitoring records can include monitoring duration and final status. Replacement records can include replacement reasons and replacement timing.

[0157] For example, query the final treatment records of each candidate historical damage from the atlas. For the top-K historical damages most similar to the current damage, calculate the percentage of each treatment method: P_repair = number of repairs / K, P_monitor = number of monitoring sessions / K, P_replace = number of replacements / K.

[0158] If P_repair > 0.6, the recommended handling strategy is "Repair Recommendation," including the average repair time and repair method for similar damage. If P_monitor > 0.7, the recommended handling strategy is "Continue Monitoring," with the recommended inspection cycle equal to the average inspection interval for similar damage.

[0159] In addition, damage development path assessment can be performed based on the data determined in the above embodiments to obtain corresponding assessment results. These results include at least a current state assessment, a summary of historical evolution, future predictions, similar case references, and comprehensive recommendations. The current state assessment can include the current depth, area, volume, percentage change from when the damage was first discovered, and percentage distance from the model threshold. The historical evolution summary can include the time of initial damage discovery, monitoring duration, average expansion rate, whether acceleration / deceleration points occurred, and the evolution pattern classification results. Future predictions can include the predicted time when the current damage reaches the threshold and an expansion trend graph (historical data points + prediction curve + confidence interval). Similar case references can include the 3-5 most similar historical damages, the final treatment results and effects of each case, and lessons learned from similar cases. Comprehensive recommendations can include treatment suggestions (e.g., repair / monitoring / replacement), suggested time points, and confidence level assessments.

[0160] The above embodiments describe the implementation method of civil aviation damage detection based on a multi-layer data storage model and the construction method of the multi-layer data storage model. Furthermore, the construction of the multi-layer data storage model also includes:

[0161] In one exemplary embodiment, to improve storage efficiency, a data processing method is provided, such as... Figure 5 As shown, it includes the following steps:

[0162] Step 502: Obtain the original data file and the original hash value of the original data file sent by the scanning device. The original data file includes grid data of the aircraft surface and three-dimensional point cloud data used to characterize the details of the aircraft surface.

[0163] The raw data file can be an unprocessed, complete binary data packet generated and sent by the scanning device, encapsulating mesh data and 3D point cloud data of the aircraft surface. The raw hash value can be a cryptographic hash value (such as SHA-256) calculated and appended by the scanning device immediately after generating the raw data file and before sending it, used to characterize a globally unique fingerprint of the file's byte sequence.

[0164] Step 504: Generate a corresponding hash value based on the data in the original data file. If the hash value is consistent with the original hash value, and / or there is a historical hash value that is the same as the original hash value, then store the original data file in the original data layer of the hierarchical storage model.

[0165] The hierarchical storage model includes a raw data layer, a feature extraction layer, a business association layer, and a graph relationship layer. The raw data layer is the lowest-level data storage unit in the hierarchical storage model, used to store historical raw data files that have passed hash verification and are unique, supporting the reproducibility verification, difference analysis, and failure rollback of all derived data in the upper layers.

[0166] In the above method, by obtaining the original data file and the original hash value of the original data file sent by the scanning device, and by judging whether the hash value and the original hash value are consistent, and judging whether there is a historical hash value with the same hash value, the original data is globally deduplicated and logically normalized. Finally, the original data file that has passed the double verification is stored in the original data layer of the hierarchical storage model, which avoids the situation of storing a large amount of damaged data with similar structures repeatedly, and improves the data storage efficiency.

[0167] To avoid the lack of versioned storage of multiple scan results of damage data and the difficulty in historical comparison, in an exemplary embodiment, the above-mentioned civil aviation damage detection method further includes: generating a version identifier for the original data file sent by the scanning device each time; associating the version identifier with the extracted data of the feature extraction layer of the hierarchical storage model, wherein the extracted data includes at least damage geometric features, location information and scan metadata.

[0168] The version identifier can be a unique identifier generated for each raw data file sent by the scanning device. As the primary key of the data hierarchy across all levels of the hierarchical storage model, the version identifier enables end-to-end traceability from the original acquisition event to derived features and then to the business entity. In essence, the raw data layer generates a version identifier for each scan of the raw file, the feature extraction layer stores the feature data of each version, and the business association layer records the repair history of the damage and associates it with all versions of scan data and repair records of the damage through the damage ID. Each layer has an independent identifier, enabling version traceability.

[0169] The feature extraction layer can be a processing layer above the original data layer in a hierarchical storage model. The extracted data can be a structured data set output by the feature extraction layer, which includes at least three core contents: damage geometric features, location information, and scan metadata. Each record is bound to a unique version identifier.

[0170] For example, after the original data file passes hash verification and is confirmed to be written to the original data layer, the scheduling service calls the version generator module to generate a unique string based on a preset template, combining the original hash value, collection timestamp, and scanning device ID. This string is then associated with the version identifier and the extracted data from the feature extraction layer of the hierarchical storage model. In this approach, both the original data layer and the feature extraction layer support data version identification, facilitating historical comparison and enabling version management and traceability.

[0171] In one exemplary embodiment, an application based on the above-described civil aviation damage detection method is provided, such as... Figure 6 The image shows the method for querying civil aviation data, which includes the following steps:

[0172] Step 602: In response to the data query request, determine the query type of the data query request.

[0173] Data query requests can be data retrieval instructions expressed in natural language or structured syntax, initiated by maintenance personnel, automated diagnostic modules, or monitoring systems. Query types can be relational graph queries or queries targeting data within the columnar storage structure of the feature extraction layer.

[0174] Step 604: Determine the target query range from the columnar storage structure and / or relational graph based on the query type.

[0175] The target query scope can be a dynamically defined logical boundary of data involved in this query execution, based on the query type. Relationship graph-based queries enable multi-hop queries, such as querying all common damage types for a specific aircraft model, or querying all damage handled by a particular technician and their corresponding aircraft. The target query scope can be at least one of a columnar storage structure and a relationship graph.

[0176] Step 606: Execute a data query request within the target query range to obtain the corresponding query data.

[0177] For example, if the user inputs "Find all instances of dent-like damage on the leading edge of the left wing of model A with a curvature anomaly > 0.8 and that have been processed by work card B, and list their 3D features, the work cards involved, and the personnel responsible" in the terminal, the system first identifies that the request contains a spatial threshold ("curvature anomaly > 0.8") and a multi-hop business relationship ("A → leading edge of left wing → dent → B chapter → work card → personnel"), determining it to be a mixed query type of "spatial positioning + business tracing". The target query range is dynamically divided into two segments: first, the set of damage identifiers with a curvature anomaly > 0.8 is filtered in the columnar storage structure, and then this set is injected into the relationship graph for querying; finally, a structured result is returned, eliminating the need for manual splicing of multiple system outputs.

[0178] Furthermore, based on the acquired query data, a defect detection report for the aircraft's damaged area can be determined. For example, an aircraft dent defect detection report can be generated, which may include basic aircraft information (such as aircraft number and model), damage information (such as damage discovery date, maintenance level, and detection model, etc.). Further, the defect detection report may also include a damage diagram related to the damage location, close-up and long-range photographs, and a parameter list. The parameter list may include a unique defect identifier, defect type, depth value, dent minor diameter, dent major diameter, distance to reference edge A, distance to reference edge B, nearest defect identifier, distance to the nearest defect, whether there was a previous maintenance operation, and a damage description, etc.

[0179] In the above embodiments, by responding to a data query request, the query type of the data query request is determined, and the target query range is determined from the columnar storage structure and / or relational graph according to the query type, thereby obtaining the query data. This eliminates the need for cross-file and database joint queries, improving query efficiency and meeting the needs of real-time interaction.

[0180] In one exemplary embodiment, such as Figure 7 The diagram illustrates a system architecture for a civil aviation damage detection method. This architecture includes a raw data layer, a feature extraction layer, a business association layer, and a graph relationship layer. The raw data layer stores the raw data files generated from 3D scanning. These files can include point cloud files, mesh files, and texture files. The 3D point cloud data in the point cloud files can include the length, width, and height of the pits, normal vectors (the pit curvature is used to calculate the pit depth), reflection intensity (different parts of the aircraft may have different materials; reflection intensity is used to identify material variations), area, and volume. The texture data in the texture files is used to enhance the realism of the 3D model and can include other attributes of the aircraft surface, such as surface roughness and metallic appearance. The data in the raw data layer uses object storage, supporting large file block storage and fast read / write operations. Each file is associated with a unique data fingerprint (data identifier) ​​for deduplication and integrity verification.

[0181] The feature extraction layer is used to extract data from the original data layer. The extracted data adopts a columnar storage structure, which supports efficient feature range query and statistics. For example, it extracts the corresponding damage geometric features and damage identifiers to determine the damage area from 3D point cloud data, and extracts the location information and scan metadata associated with the damage area from mesh data. The extracted data are stored in a columnar storage structure.

[0182] The business relationship layer uses a relational database to store business logic relationship information, including multiple target data tables associated with aircraft maintenance business data. These target data tables include a damage table, an aircraft type table, a maintenance record table, and a manual chapter table. Many-to-many relationships between these target data tables are established through foreign key relationships, creating a relationship between damage, aircraft type, record, and manual. The graph relationship layer uses a graph database to store complex relationships between entities.

[0183] For example, based on the system architecture diagram above, scan data of the aircraft surface is uploaded to the raw data layer and stored as files, generating a data fingerprint for each file. The feature extraction service processes the raw data, extracting data from the raw data layer and writing it into the feature extraction layer. The business system, based on maintenance task-related information such as aircraft model and work card, writes this information into the business association layer, establishing many-to-many relationships between damage, aircraft model, record, and manual through foreign key relationships. The graph construction service synchronously updates the graph relationship layer, establishing entity relationship edges. Its specific implementation can be achieved as described above and will not be elaborated upon here.

[0184] In one exemplary embodiment, such as Figure 8 As shown, a data writing sequence diagram is provided, specifically including: the scanning device uploads point cloud data to the raw data layer; the raw data layer stores the uploaded scanning data and returns a unique storage ID and data fingerprint; the feature extraction service requests raw data from the raw data layer; the raw data layer returns point cloud / mesh data; the feature extraction service extracts data from the data returned by the raw data layer; the feature extraction service performs feature calculations and stores them in the feature extraction layer, returns a feature ID, and returns the feature ID to the caller; the business system associates the feature ID with business data such as device model and work card, and writes it to the business association layer; the business association layer returns the association record ID; the graph construction service updates the graph relationship layer based on the business association data, and the entire writing process is completed.

[0185] It should be noted that the specific implementation of this example can be achieved in the manner described above, and will not be repeated here.

[0186] In the above embodiments, a four-layer storage model consisting of "raw data layer - feature extraction layer - business association layer - graph relationship layer" is constructed to achieve deep integration and efficient management of aircraft 3D damage data and maintenance business data. This four-layer storage model enables rapid retrieval of damage data, version tracking, relationship graph querying, and dynamic expansion, significantly improving data query efficiency and system maintainability.

[0187] In one exemplary example, a civil aviation damage detection method is provided, such as... Figure 9 As shown, it includes the following:

[0188] A multi-layered data storage model integrating aircraft damage data and maintenance operational data is constructed. This model includes at least a feature extraction layer, a graph relationship layer, and a business association layer. Three-dimensional scan data of the target aircraft is acquired, comprising mesh data and 3D point cloud data. Point cloud preprocessing is performed on the 3D scan data, and a mesh model is constructed based on the mesh data. The 3D point cloud data and mesh data are analyzed to extract the damage feature vector of the current damage and determine its damage identifier. Damage identification is performed based on the damage feature vector. If the current damage type is a known damage type, a matching historical damage dataset is retrieved from the multi-layered data storage model, and the damage evolution record table is updated based on this dataset. If the damage type is not known, a new damage record is created, and the loss feature vector is normalized before retrieving the matching historical damage dataset from the multi-layered data storage model.

[0189] For each historical damage in the historical damage dataset, determine the weighted similarity between the historical damage feature vector and the damage feature vector of each historical damage; based on the weighted similarity, determine the candidate historical damage dataset from the historical damage dataset; perform voting statistics on the damage type corresponding to each candidate historical damage in the candidate historical damage dataset, and if the determined damage type does not match the current damage type, correct the current damage type and determine the damage type of the current damage.

[0190] For each candidate historical damage in the candidate historical damage dataset, historical scan records are extracted from a multi-layer data storage model. These historical scan records are time-series data. For each time-series data segment, the data is divided into segments, and the expansion rate of each segment is calculated to obtain an expansion rate sequence. Based on the expansion rate sequence and the time-series data segments, points of change in expansion patterns are identified. Based on these points and the expansion rate sequence, a first damage evolution pattern for each candidate historical damage is determined. From the determined first damage evolution pattern, a second damage evolution pattern matching the damage feature vector is identified, and a target prediction model matching the second damage evolution pattern is obtained. The second damage evolution pattern and the damage feature vector are input into the target prediction model, outputting the predicted damage depth and prediction confidence interval for the target aircraft. Furthermore, a recommended processing strategy for the current damage and a damage development path assessment can be generated. A final detection report is generated based on the detection data obtained in the preceding steps and then visualized.

[0191] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0192] Based on the same inventive concept, this application also provides a civil aviation damage detection device for implementing the aforementioned civil aviation damage detection method. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the civil aviation damage detection device provided below can be found in the limitations of the civil aviation damage detection method described above, and will not be repeated here.

[0193] In one exemplary embodiment, such as Figure 10 As shown, a civil aviation damage detection device is provided, comprising: a model building module 1002, a data acquisition module 1004, a map query module 1006, and a damage detection module 1008, wherein:

[0194] The model building module 1002 is used to build a multi-layer data storage model that integrates aircraft damage data and maintenance business data; the multi-layer data storage model includes at least a feature extraction layer, a graph relationship layer, and a business association layer.

[0195] The data acquisition module 1004 is used to acquire the damage feature vector of the current damage to the target aircraft.

[0196] The graph query module 1006 is used to query the relation graph of the graph relation layer based on the target aircraft and the damage feature vector if the damage feature vector does not meet the preset damage conditions, determine the corresponding similar damage identifier, and obtain the corresponding historical damage dataset based on the similar damage identifier.

[0197] The damage detection module 1008 is used to perform damage detection on the target aircraft based on historical damage datasets and determine the damage detection results of the target aircraft.

[0198] The aforementioned civil aviation damage detection device constructs a multi-layered data storage model that integrates aircraft damage data and maintenance business data. Based on this multi-layered data storage model, when performing damage detection on a target aircraft, it obtains the damage feature vector of the target aircraft from the feature extraction layer according to the damage identifier carried by the damage detection command. This eliminates the need for cross-system and cross-database data queries. The data storage model based on this storage format can quickly obtain the damage feature vector corresponding to the damage identifier. Furthermore, if the damage feature vector does not meet the preset damage conditions, a relational graph is introduced to further assist in the aircraft damage detection. That is, based on the target aircraft and the damage feature vector, the relational graph of the graph relation layer is queried to determine the corresponding similar damage identifier. Based on the similar damage identifier, the corresponding historical damage dataset is obtained. By performing damage confidence enhancement detection on the target aircraft based on the historical damage dataset, the damage detection result of the target aircraft is determined. The entire process replaces the discrete experience comparison after manual measurement, avoids the time-consuming and laborious problem of manual processing, and improves the efficiency of civil aviation damage detection.

[0199] In an exemplary embodiment, the model building module 1002 is used to acquire mesh data of the aircraft surface and three-dimensional point cloud data for characterizing the details of the aircraft surface.

[0200] Analyze 3D point cloud data and mesh data to extract aircraft damage data of damaged areas from 3D point cloud data and determine damage indicators of damaged areas. Aircraft damage data includes damage geometric features and texture features.

[0201] Extract location information and scan metadata associated with the damaged area from the grid data;

[0202] The aircraft damage data and scan metadata of the damaged area are stored in the feature extraction layer in a columnar storage structure, and the corresponding damage identifier is returned.

[0203] Obtain multiple target data tables associated with aircraft maintenance business data, and establish the relationship between the target data tables through foreign key relationships at the business association layer to link aircraft damage data and aircraft maintenance business data; the damage table in the target data table is associated with damage identifiers;

[0204] Entity identification is performed on multiple target data tables. By determining the relationships between entities corresponding to multiple target data tables, a relationship graph is constructed at the graph relationship layer to obtain a multi-layer data storage model.

[0205] In an exemplary embodiment, the above-described apparatus further includes a relational graph update module, used to perform entity recognition on multiple target data tables, determine the entities corresponding to the multiple target data tables and the relationships between the entities; the entities include at least aircraft, damage, parts, work cards, personnel and manual chapters;

[0206] Entities are mapped to nodes in the graph, and edges are determined to connect their respective nodes based on the relationships. A relation graph is then constructed in the graph relation layer based on the nodes and edges.

[0207] In an exemplary embodiment, the atlas query module 1006 is used to determine that the damage feature vector does not meet the preset damage condition if the geometric feature does not meet the preset geometric feature condition but the curvature feature meets the preset curvature condition.

[0208] The first historical damage node matched with the target aircraft in the relation graph of the relation layer, and the second historical damage node matched with the damage feature vector were determined.

[0209] Based on the first and second historical damage nodes, the corresponding damage data is obtained from the multi-layer data storage model to determine the historical damage dataset; the damage data corresponding to each historical damage node includes instance information of a complete historical damage.

[0210] In an exemplary embodiment, the map query module 1006 is used to determine the damage feature vector at the damage location corresponding to the target aircraft, and the aircraft identifier of the target aircraft.

[0211] Based on the damage location and / or aircraft identification, determine the first matching historical damage node from the relation graph of the relation layer;

[0212] Calculate the similarity between the historical damage feature vector and the damage feature vector corresponding to the historical damage node in the relationship graph, and determine the historical damage node whose similarity is within the preset similarity range as the second historical damage node.

[0213] In an exemplary embodiment, the damage detection module 1008 is used to determine the weighted similarity between the historical damage feature vector and the damage feature vector for each historical damage in the historical damage dataset.

[0214] Based on weighted similarity, candidate historical damage datasets are determined from the historical damage dataset; for each candidate historical damage in the candidate historical damage dataset, the damage type corresponding to each candidate historical damage is voted on and statistically analyzed to determine the damage type of the current damage.

[0215] In an exemplary embodiment, the damage detection module 1008 is used to perform time series evolution analysis processing on each candidate historical damage in the candidate historical damage dataset based on the historical scan record data of each candidate historical damage, and determine a second damage evolution pattern that matches the damage feature vector.

[0216] The damage feature vector is predicted based on the second damage evolution model to obtain damage trend prediction results and prediction confidence data.

[0217] In an exemplary embodiment, the damage detection module 1008 is used to extract historical scan record data of each candidate historical damage from a multi-layer data storage model for each candidate historical damage in the candidate historical damage dataset; the historical scan record data is time series data.

[0218] For each time series data, the time series data is segmented, and the spread rate of each time series data segment is calculated to obtain the spread rate sequence;

[0219] Identify points of change in the expansion pattern based on the expansion rate sequence and time series data segments;

[0220] Based on the expansion pattern change points and expansion rate sequence, the first damage evolution pattern of each candidate historical damage is determined, and the second damage evolution pattern that matches the damage feature vector is determined from the determined first damage evolution pattern.

[0221] In an exemplary embodiment, the damage detection module 1008 is used to determine the damage prediction error corresponding to each time point in the time series data segment in turn, and to determine the candidate expansion mode change point by performing cumulative deviation processing on the damage prediction error.

[0222] The candidate expansion mode change point is determined based on the expansion rate sequence, and the ratio of the expansion rate before and after the change is obtained.

[0223] The candidate expansion mode change points are verified and the expansion modes are classified based on the expansion rate ratio to determine the expansion mode change points.

[0224] In an exemplary embodiment, the damage detection module 1008 is used to construct an evolution feature vector based on the expansion rate sequence and expansion pattern change points;

[0225] Clustering is performed on the evolution feature vectors to determine the first damage evolution pattern corresponding to each candidate historical damage and the class center feature corresponding to each first damage evolution pattern.

[0226] Calculate the distance between the damage feature vector and various central features, and determine the second damage evolution pattern that matches the damage feature vector from the first damage evolution pattern based on the distance.

[0227] In an exemplary embodiment, the damage detection module 1008 is used to acquire a target prediction model that matches the second damage evolution pattern.

[0228] The second damage evolution mode and damage feature vector are input into the target prediction model, and the predicted damage depth and prediction confidence interval of the target aircraft are output.

[0229] The aforementioned civil aviation damage detection device extracts data from raw data acquired from the aircraft surface, including mesh data and 3D point cloud data used to characterize aircraft surface details. It stores the extracted damage feature data, associated location information, and scan metadata in a columnar storage structure. This data is then linked to damage tables in multiple target data tables associated with aircraft maintenance business data through corresponding damage identifiers. Furthermore, foreign key relationships are established between these target data tables to link 3D damage data and aircraft maintenance business data. This addresses the lack of structured relationships between data in traditional storage methods. This storage approach shortens the response time of complex queries using multi-layered index structures, improving query efficiency and avoiding slow responses to cross-file and database queries that fail to meet real-time interaction requirements. Moreover, this data storage method supports dynamically adding damage attributes and business relationships, adapting to business changes.

[0230] Each module in the aforementioned civil aviation damage detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0231] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 11As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a civil aviation damage detection method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0232] Those skilled in the art will understand that Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0233] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0234] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0235] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0236] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0237] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0238] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0239] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for detecting damage in civil aviation aircraft, characterized in that, include: A multi-layered data storage model integrating aircraft damage data and maintenance business data is constructed; the multi-layered data storage model includes at least a feature extraction layer, a graph relationship layer, and a business association layer. Obtain the damage feature vector of the target aircraft's current damage; If the damage feature vector does not meet the preset damage conditions, then the relationship graph of the graph relationship layer is queried according to the target aircraft and the damage feature vector to determine the corresponding similar damage identifier, and the corresponding historical damage dataset is obtained according to the similar damage identifier. Damage detection is performed on the target aircraft based on the historical damage dataset to determine the damage detection results of the target aircraft.

2. The method according to claim 1, characterized in that, The construction of a multi-layered data storage model that integrates aircraft damage data and maintenance operational data includes: Acquire mesh data of the aircraft surface and 3D point cloud data for characterizing the details of the aircraft surface; The three-dimensional point cloud data and the mesh data are analyzed to extract aircraft damage data of the damaged area from the three-dimensional point cloud data and to determine the damage identifier of the damaged area. Extract location information and scan metadata associated with the damaged area from the grid data; The aircraft damage data and the scan metadata of the damaged area are stored in the feature extraction layer in a columnar storage structure, and the corresponding damage identifier is returned. Multiple target data tables associated with aircraft maintenance business data are obtained. In the business association layer, the association relationship between the target data tables is established through foreign key relationships to associate the aircraft damage data and the aircraft maintenance business data; the damage table in the target data table is associated with the damage identifier. Entity identification is performed on multiple target data tables. By determining the relationships between entities corresponding to the multiple target data tables, a relationship graph is constructed in the graph relationship layer to obtain a multi-layer data storage model.

3. The method according to claim 2, characterized in that, The step of performing entity recognition on multiple target data tables, and constructing a relationship graph in the graph relationship layer by determining the association relationships between entities corresponding to the multiple target data tables, includes: Entity identification is performed on multiple target data tables to determine the entities corresponding to the multiple target data tables and the relationships between the entities; the entities include at least aircraft, damage, components, work cards, personnel, and manual chapters. The entities are mapped to nodes in the graph, and edges for connecting their respective nodes are determined based on the association relationships. A relation graph is constructed in the graph relation layer based on the nodes and the edges.

4. The method according to claim 1 or 2, characterized in that, The damage feature vector includes at least geometric features, curvature features, and texture features. If the damage feature vector does not meet the preset damage conditions, a query is performed from the relational graph of the graph relation layer based on the target aircraft and the damage feature vector to determine the corresponding similar damage identifier. Based on the similar damage identifier, the corresponding historical damage dataset is obtained, including: If the geometric feature does not satisfy the preset geometric feature condition but the curvature feature satisfies the preset curvature condition, then it is determined that the damage feature vector does not satisfy the preset damage condition. The first historical damage node that matches the target aircraft in the relation graph of the relation layer, and the second historical damage node that matches the damage feature vector are determined. Based on the first historical damage node and the second historical damage node, the corresponding damage data is obtained from the multi-layer data storage model to determine the historical damage dataset; the damage data corresponding to each historical damage node includes instance information of a complete historical damage.

5. The method according to claim 4, characterized in that, The process of determining the first historical damage node matched with the target aircraft from the relation graph of the relation layer, and the second historical damage node matched with the damage feature vector, includes: Determine the damage location of the damage feature vector on the target aircraft, and the aircraft identifier of the target aircraft; Based on the damage location and / or the aircraft identifier, determine the first matching historical damage node from the relation graph of the relation layer; Calculate the similarity between the historical damage feature vector corresponding to the historical damage node in the relationship graph and the damage feature vector, and determine the historical damage node whose similarity is within a preset similarity range as the second historical damage node.

6. The method according to claim 1 or 2, characterized in that, The step of performing damage detection on the target aircraft based on the historical damage dataset and determining the damage detection result of the target aircraft includes: For each historical damage in the historical damage dataset, determine the weighted similarity between the historical damage feature vector and the damage feature vector of each historical damage. Based on the weighted similarity, candidate historical damage datasets are determined from the historical damage dataset; The damage type of the current damage is determined by voting statistics on the damage type corresponding to each candidate historical damage in the candidate historical damage dataset.

7. The method according to claim 6, characterized in that, The method further includes: For each candidate historical damage in the candidate historical damage dataset, time series evolution analysis is performed based on the historical scan record data of each candidate historical damage to determine a second damage evolution pattern that matches the damage feature vector. The damage feature vector is predicted based on the second damage evolution pattern to obtain damage trend prediction results and prediction confidence data.

8. The method according to claim 6, characterized in that, For each candidate historical damage in the candidate historical damage dataset, time series evolution analysis is performed based on the historical scan records of each candidate historical damage to determine a second damage evolution pattern that matches the damage feature vector, including: For each candidate historical damage in the candidate historical damage dataset, historical scan record data for each candidate historical damage is extracted from the multi-layer data storage model; the historical scan record data is time series data. For each of the time series data, the time series data is segmented, and the expansion rate of each time series data segment is calculated to obtain an expansion rate sequence; Based on the expansion rate sequence and the time series data segment, identify the points of change in the expansion pattern; Based on the expansion pattern change points and the expansion rate sequence, a first damage evolution pattern is determined for each candidate historical damage, and a second damage evolution pattern matching the damage feature vector is determined from the determined first damage evolution patterns.

9. The method according to claim 8, characterized in that, The step of identifying points of change in the expansion pattern based on the expansion rate sequence and the time series data segment includes: For each time point in the time series data segment, the damage prediction error corresponding to each time point is determined sequentially. By performing cumulative deviation processing on the damage prediction error, candidate expansion mode change points are determined. The candidate expansion mode change point is determined based on the expansion rate sequence, and the expansion rate ratio before and after the change is calculated. The candidate expansion mode change points are verified and the expansion modes are classified based on the expansion rate ratio to determine the expansion mode change points.

10. The method according to claim 8, characterized in that, The step of determining a first damage evolution pattern for each candidate historical damage based on the expansion pattern change points and the expansion rate sequence, and determining a second damage evolution pattern matching the damage feature vector from the determined first damage evolution patterns, includes: Based on the expansion rate sequence and the expansion mode change points, an evolution feature vector is constructed; Clustering is performed on the evolution feature vector to determine the first damage evolution pattern corresponding to the candidate historical damage and the class center feature corresponding to each first damage evolution pattern. Calculate the distance between the damage feature vector and each of the class center features, and determine a second damage evolution pattern that matches the damage feature vector from the first damage evolution pattern based on the distance.

11. The method according to claim 7, characterized in that, The step of predicting the damage feature vector based on the second damage evolution pattern to obtain damage trend prediction results and prediction confidence data includes: Obtain a target prediction model that matches the second damage evolution pattern; The second damage evolution pattern and the damage feature vector are input into the target prediction model, and the predicted damage depth and prediction confidence interval of the target aircraft are output.

12. A civil aviation damage detection device, characterized in that, include: The model building module is used to construct a multi-layer data storage model that integrates aircraft damage data and maintenance business data; the multi-layer data storage model includes at least a feature extraction layer, a graph relationship layer, and a business association layer. The data acquisition module is used to acquire the damage feature vector of the current damage to the target aircraft; The graph query module is used to query the relationship graph of the graph relationship layer based on the target aircraft and the damage feature vector if the damage feature vector does not meet the preset damage conditions, determine the corresponding similar damage identifier, and obtain the corresponding historical damage dataset based on the similar damage identifier. The damage detection module is used to perform damage detection on the target aircraft based on the historical damage dataset and determine the damage detection result of the target aircraft.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 11.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 11.

Images