An artificial intelligence-based aviation material management method and system

Abstract

The application discloses an artificial intelligence-based aviation material management method and system, relates to the technical field of aviation material management, and comprises the following steps: acquiring a unique part number code stored in an electronic tag of aviation material to be taken and a taking visual image of the aviation material to be taken; based on the unique part number code, a corresponding standard visual feature template is searched from a pre-constructed standard feature library; a visual feature vector of the taking visual image is extracted, and a first similarity is calculated; when the first similarity meets a first preset condition, it is determined that the taking check passes; in the assembly link, a target installation position identifier contained in a current electronic maintenance work order and an assembly visual image of the aviation material to be assembled are acquired; a visual feature vector of the assembly visual image is extracted, and a second similarity is calculated; and when the second similarity meets a second preset condition, an assembly permission instruction is output. The application has the effect of improving aviation material management efficiency.

Description

Technical Field

[0001] This application relates to the field of aircraft material management technology, and in particular to an artificial intelligence-based aircraft material management method and system. Background Technology

[0002] In the field of aircraft material management, ensuring absolute accuracy in the requisition and assembly of aircraft materials is the core of guaranteeing flight safety and maintenance quality. Traditional management methods have significant shortcomings in error prevention, relying mainly on human experience and decentralized automation tools, which are difficult to adapt to the stringent requirements of modern aircraft maintenance for high efficiency and high reliability. The potential risks arising from this have become a long-standing pain point of concern in the industry.

[0003] Existing technologies typically use barcodes or RFID tags as the primary identification for aircraft materials. While this method achieves a certain degree of automated information entry, its error prevention capabilities have fundamental limitations. First, the tags themselves may be damaged, detached, or have incorrect information written, making the data unreliable from the source. Second, and more critically, the binding relationship between the physical aircraft material and its electronic tag is fragile. Even if the system successfully reads a correct tag, it cannot verify whether the physical component being received or installed is the correct aircraft material corresponding to that tag. When faced with aircraft materials that look similar but have vastly different models and specifications, simply relying on tag scanning cannot prevent human errors such as misattribution. Furthermore, existing visual-assisted identification technologies are often independent of the main process, used only for simple information display or recording, and fail to achieve deep, logically rigorous real-time verification and integration with core requisition and assembly instructions and the electronic identification of aircraft materials, thus limiting their application effectiveness.

[0004] Furthermore, the data and verification actions at each stage of the existing management process are isolated, failing to form a coherent closed loop. The verification requirements at the requisition stage are disconnected from those at the assembly stage; the system lacks the ability to automatically compare outbound aircraft material information with the installation location requirements in specific maintenance work orders. This results in a lack of a final verification mechanism based on the work order environment before final assembly. Errors occurring at the requisition stage can easily propagate to the installation stage, creating potential safety hazards. Simultaneously, existing methods are typically static and rigid, unable to learn and optimize their judgment criteria through continuous operation. For example, they cannot adapt to reasonable changes in the appearance of aircraft materials due to normal use, potentially leading to false alarms or missed alarms, affecting the system's usability and reliability.

[0005] To address the aforementioned issues, there is an urgent need in this field for an intelligent management method that can span the key stages of aircraft material requisition and assembly, achieve dual binding and verification of electronic identity and physical entity, and be deeply integrated with maintenance tasks, in order to build a closed-loop, reliable, and adaptive error-proofing system and fundamentally improve the safety and accuracy of aircraft material management. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this application provides an artificial intelligence-based method and system for managing aviation materials.

[0007] In a first aspect, this application provides an artificial intelligence-based aircraft material management method, comprising the following steps:

[0008] The unique part number code stored in the electronic tag of the aviation material to be requisitioned is obtained by the radio frequency identification reader, and the visual image of the aviation material to be requisitioned is captured by the first image acquisition device.

[0009] Based on the unique part number code, the corresponding standard visual feature template is retrieved from the pre-built standard feature library;

[0010] Extract the visual feature vector of the requisition visual image, and calculate the first similarity between the visual feature vector and the standard visual feature template;

[0011] When the first similarity meets the first preset condition, the requisition verification is determined to be successful, and the unique part number is associated with the current requisition task.

[0012] During the assembly process, the target installation location identifier contained in the current electronic maintenance work order is obtained, as well as the assembly visual image of the aircraft material to be assembled captured by the second image acquisition device.

[0013] Extract the visual feature vector of the assembly visual image, and calculate the second similarity between the visual feature vector of the assembly visual image and the standard visual feature template corresponding to the associated unique part number code;

[0014] When the second similarity meets the second preset condition, an assembly permission command is output.

[0015] Preferably, the construction steps of the pre-built standard feature library specifically include:

[0016] Acquire multi-angle sample images of various aviation materials under standard conditions;

[0017] For various types of aviation materials, a feature extraction network is used to process multi-angle sample images corresponding to various types of aviation materials to generate a standard visual feature vector set for the various types of aviation materials.

[0018] The standard visual feature vector set is associated with and stored with the corresponding unique part number code to form a standard visual feature template.

[0019] Preferably, extracting the visual feature vector of the requisition visual image specifically includes:

[0020] The requisition visual image is input into a pre-trained feature extraction network;

[0021] The feature vector output by the feature extraction network at a specific layer is obtained and used as the visual feature vector of the visual image being used.

[0022] The feature extraction network has the same network structure as the network used to construct the standard visual feature template.

[0023] Preferably, when the first similarity meets the first preset condition, the requisition verification is determined to be passed, including:

[0024] Compare the first similarity with the first preset threshold;

[0025] If the first similarity is greater than or equal to the first preset threshold, it is determined that the first preset condition is met, and a first consistency identifier is generated;

[0026] If the first similarity is less than the first preset threshold, it is determined that the first preset condition is not met, a first alarm command is generated, and the requisition process is interrupted.

[0027] Preferred options also include:

[0028] During the assembly process, an image of the on-site environment at the target installation location is acquired by a third image acquisition device;

[0029] Based on the target installation location identifier, retrieve the corresponding standard installation location feature template from the standard feature library;

[0030] Extract the visual feature vector of the on-site environment image, and calculate the third similarity between the visual feature vector of the on-site environment image and the standard installation position feature template;

[0031] The output assembly license instruction must be executed when the third similarity meets the third preset condition.

[0032] Preferably, the output assembly permission instruction includes:

[0033] When the second similarity satisfies the second preset condition and the third similarity satisfies the third preset condition, a second consistency identifier is generated;

[0034] The second consistency identifier, the unique part number code, and the target installation location identifier are associated and recorded in the electronic maintenance record.

[0035] Preferably, when the requisition verification fails, the following steps are also included:

[0036] Obtain one or more candidate part number codes that are visually similar to the currently read unique part number code;

[0037] Obtain the candidate standard visual feature template corresponding to the candidate part number code from the standard feature library;

[0038] Calculate the candidate similarity between the visual feature vector of the requisition visual image and each of the candidate standard visual feature templates;

[0039] The output includes a prompt message containing the candidate part number code and its corresponding candidate similarity.

[0040] Preferably, it also includes a historical data analysis step:

[0041] The failure records and final fault confirmation records of each unique part number code corresponding to the aircraft material during the requisition verification and assembly verification stages are statistically analyzed.

[0042] Based on the statistical results, the matching threshold of the standard visual feature templates for the corresponding aircraft materials is adjusted.

[0043] Secondly, this application provides an artificial intelligence-based aircraft material management system, comprising:

[0044] The first data acquisition module is used to acquire the unique part number code stored in the electronic tag of the aviation material to be requisitioned, which is read by the radio frequency identification reader, and the requisition visual image of the aviation material to be requisitioned, which is captured by the first image acquisition device.

[0045] The retrieval module is used to retrieve the corresponding standard visual feature template from the pre-built standard feature library based on the unique part number code;

[0046] The first similarity calculation module is used to extract the visual feature vector of the requisition visual image and calculate the first similarity between the visual feature vector and the standard visual feature template.

[0047] The requisition verification and determination module is used to determine that the requisition verification is passed when the first similarity meets the first preset condition, and to associate the unique item number code with the current requisition task;

[0048] The second data acquisition module is used to acquire, during the assembly process, the target installation location identifier contained in the current electronic maintenance work order, as well as the assembly visual image of the aircraft material to be assembled captured by the second image acquisition device.

[0049] The second similarity calculation module is used to extract the visual feature vector of the assembly visual image and calculate the second similarity between the visual feature vector of the assembly visual image and the standard visual feature template corresponding to the associated unique part number code.

[0050] The instruction output module is used to output an assembly permission instruction when the second similarity meets the second preset condition.

[0051] Thirdly, this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform any of the above-described artificial intelligence-based aircraft material management methods.

[0052] In summary, this application includes at least one of the following beneficial technical effects:

[0053] 1. This application provides an artificial intelligence-based method for managing aircraft materials. This invention breaks through the limitations of traditional methods that rely solely on electronic identification. It creatively introduces visual feature comparison in the requisition process, realizing mandatory dual binding and real-time verification of the unique part number code of the aircraft material and the visual image of the requisition. This effectively solves the problems of unreliable source data caused by damaged identification objects, incorrect information, or separation of the object and the identification, as well as the problem of difficulty in distinguishing similar-looking parts. It can proactively intercept aircraft materials with inconsistent identities at the source, transforming the passive error prevention that relies on manual experience and single-point verification into proactive and accurate error prevention based on multi-source data fusion. This greatly reduces the risk of requisitioning the wrong parts and lays a solid safety foundation for subsequent processes.

[0054] 2. By intelligently linking the visual verification of the assembly process with the verification results of the requisition process and the electronic maintenance work order, a closed-loop management system is achieved for the entire process from warehouse outbound to on-site assembly. This not only reconfirms the identity of aircraft materials before assembly to prevent substitution during the transfer process, but also innovatively introduces environmental visual verification of the target installation location to ensure that the "correct parts" are installed in the "correct locations". This effectively changes the situation of isolated data and disconnected verification in each process, forming an interlocking verification chain. This allows the system to automatically detect and block errors that occur in the requisition process or confusion in the assembly process, effectively preventing the transmission of errors to the installation stage.

[0055] 3. This invention is not a static rule system, but possesses dynamic optimization and intelligent assistance capabilities. On the one hand, it can adaptively adjust the matching threshold through historical data analysis and optimize feature templates using successful verification data, enabling the system to adapt to normal appearance changes of aircraft materials, reducing false alarms and missed alarms, and solving the problems of system rigidity and insufficient practicality in the background technology. On the other hand, when the requisition verification fails, the system can proactively provide a list of visually similar candidate part numbers, upgrading a simple alarm to an efficient error correction assistance, significantly improving the handling efficiency and human-machine collaboration experience in abnormal situations. In addition, all key verification and authorization operations automatically generate tamper-proof digital records, realizing full-process traceability of maintenance quality and providing data support for safety management and continuous improvement. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart of an artificial intelligence-based aircraft material management method according to an embodiment of this application.

[0058] Figure 2 This is a schematic diagram of an artificial intelligence-based aircraft material management system according to an embodiment of this application. Detailed Implementation

[0059] The following description, in conjunction with the implementation of this invention, is merely an example and illustration of the concept of this invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the inventive concept or exceed the scope defined in these claims, all of which should fall within the protection scope of this invention.

[0060] Overview of this application:

[0061] In existing technologies, error prevention management in the requisition and assembly of aircraft materials mainly relies on single electronic tags such as barcodes or RFID tags, as well as manual visual verification. This method has inherent defects: electronic tags are prone to damage, detachment, or information errors, leading to inaccurate data sources, and cannot verify the consistency between physical entities and electronic tags; when faced with different aircraft materials that look similar, simply scanning the tags cannot prevent human errors such as mislabeling. In addition, visual assistance technologies are often independent of core processes, used only for information display, and fail to achieve deep integration and real-time verification with requisition and assembly instructions. Data and verification actions at each stage are isolated from each other, requisition verification is disconnected from assembly requirements, and there is a lack of a closed-loop review mechanism based on maintenance work orders. Once an error occurs in an early stage, it may be transmitted to the final installation, creating potential safety hazards. At the same time, existing methods are static and rigid, unable to learn from historical operations to adapt to normal changes in the appearance of aircraft materials, and are prone to false alarms or missed alarms.

[0062] To address the aforementioned issues, the inventors discovered that the dual binding of electronic identification data and physical visual features of aircraft materials, along with closed-loop verification across various stages, is crucial for building a reliable error-proofing system. During the research, it was found that while the unique part number code provided by RFID serves as a standardized "electronic identity," it is easily separated from the physical entity; and while visual features uniquely characterize the physical appearance, they are susceptible to interference from lighting and angle. By forcibly linking and verifying both during the requisition process, the authenticity of the issued entity's identity can be ensured. Furthermore, the inventors recognized the need to incorporate maintenance task context into the assembly process. By comparing the visual features of the parts to be assembled with the target installation location, the correctness of the assembly position can be verified, thus forming a complete verification chain from "warehouse" to "workstation." Based on this, by establishing an adaptive update mechanism for standard feature templates, the system can tolerate reasonable changes in the appearance of aircraft materials, improving practicality and reliability.

[0063] Specifically, the detection system first simultaneously acquires the unique part number code of the aircraft material read by the RFID reader and the visual image of the requisition captured by the image acquisition device during the requisition process. The system uses this code to retrieve a pre-constructed standard visual feature template and extracts the visual feature vector of the requisition image for similarity calculation. If preset conditions are met, the verification passes, and the code is associated with the current task. During the assembly process, the system acquires the target installation location identifier from the current electronic maintenance work order, the assembly visual image of the aircraft material to be installed, and the on-site environmental image of the target installation location. The system performs a dual match: first, it matches the assembly image features with the standard template corresponding to the associated code to verify that the aircraft material's identity has not changed during circulation; second, it matches the on-site environmental image features with the standard installation location template corresponding to the location identifier to verify that the assembly location is correct. Only when both matches are successful does the system output an assembly permission instruction. The system also includes a closed-loop optimization step, dynamically updating the standard feature template using the successfully verified image data to adapt to actual working conditions.

[0064] Compared to existing technologies, traditional methods rely on a single data source and operate in isolated stages, failing to establish a reliable binding between physical and electronic identities, and lacking end-to-end verification based on maintenance tasks. This solution innovatively integrates RFID and multi-view visual data, constructing a cross-stage intelligent error-prevention closed loop through "identity-feature" binding verification in the requisition stage and "identity-location" dual verification in the assembly stage. Unlike existing static and independent verification methods, this solution achieves deep integration of data flow and business flow, and can adaptively optimize feature templates through a feedback mechanism, significantly improving the accuracy and robustness of aircraft material identification in complex real-world scenarios.

[0065] Through the above technical solution, this application effectively overcomes the problems of errors in the requisition and assembly of aircraft materials caused by incorrect identification, human negligence, or similar appearance, and greatly improves the reliability of verification while maintaining automation and efficiency. The dynamic dual verification mechanism takes into account both the uniqueness of electronic identity and the physical nature of visual features, and the closed-loop adaptive function ensures the stability and accuracy of the system in long-term operation.

[0066] Example 1

[0067] This application discloses an artificial intelligence-based method for managing aviation materials.

[0068] Reference Figure 1 An artificial intelligence-based aircraft material management method includes the following steps:

[0069] The unique part number code stored in the electronic tag of the aviation material to be requisitioned is obtained by the radio frequency identification reader, and the visual image of the aviation material to be requisitioned is captured by the first image acquisition device.

[0070] Based on the unique part number code, the corresponding standard visual feature template is retrieved from the pre-built standard feature library;

[0071] Extract the visual feature vector of the requisition visual image, and calculate the first similarity between the visual feature vector and the standard visual feature template;

[0072] When the first similarity meets the first preset condition, the requisition verification is determined to be successful, and the unique part number is associated with the current requisition task.

[0073] During the assembly process, the target installation location identifier contained in the current electronic maintenance work order is obtained, as well as the assembly visual image of the aircraft material to be assembled captured by the second image acquisition device.

[0074] Extract the visual feature vector of the assembly visual image, and calculate the second similarity between the visual feature vector of the assembly visual image and the standard visual feature template corresponding to the associated unique part number code;

[0075] When the second similarity meets the second preset condition, an assembly permission command is output.

[0076] Specifically, the acquisition of the unique part number code stored in the electronic tag of the aircraft material to be issued, which is read by the radio frequency identification reader, refers to the non-contact reading of the identity information stored in the electronic tag attached to the aircraft material through radio frequency identification technology, which is used as the unique digital identity certificate for the aircraft material to be retrieved and tracked in the information system.

[0077] The acquisition of the visual image of the aircraft materials to be requisitioned, captured by the first image acquisition device, refers to the acquisition of a two-dimensional digital image of the aircraft materials to be requisitioned at a fixed workstation at the material requisition window of the aircraft material warehouse, using an industrial camera or high-definition camera under preset lighting conditions. Specifically, it can be achieved by using an area array CCD or CMOS sensor in conjunction with a fixed focal length lens, which is used to acquire the appearance visual information of the physical entity of the aircraft materials, as a visual basis for verifying consistency with electronic identity information.

[0078] The step of retrieving the corresponding standard visual feature template from the pre-built standard feature library based on the unique part number code refers to retrieving the standard visual feature data pre-established for this type of aircraft material from the local or cloud database using the read unique part number code as the query key. Specifically, it can be achieved by using the index query or key-value pair storage retrieval of a relational database. The standard visual feature template serves as the benchmark reference for subsequent visual similarity comparison.

[0079] The extraction of the visual feature vector of the requisition visual image refers to the extraction of high-dimensional digital features that can characterize the essential attributes of the appearance from the requisition visual image using computer vision algorithms. Specifically, it can be achieved by using a feature extractor based on a deep convolutional neural network or a traditionally hand-designed feature descriptor, which is used to convert image information into a numerical representation that can be mathematically measured.

[0080] The calculation of the first similarity between the visual feature vector and the standard visual feature template refers to using a specific distance or similarity metric function to numerically compare the extracted real-time visual feature vector with the retrieved standard visual feature template. Specifically, it can be implemented using cosine similarity, Euclidean distance, or Mahalanobis distance to quantify the degree of matching between the appearance of the currently issued aircraft materials and the standard sample. In a specific embodiment of the present invention, the calculation of the first similarity is achieved by measuring the directional consistency between the visual feature vector of the issued visual image and the standard visual feature template in a high-dimensional feature space.

[0081] Specifically, the calculation process can be expressed by the following mathematical formula:

[0082] Let the extracted visual feature vector of the requisition visual image be... Let the feature vector of the corresponding standard visual feature template retrieved from the standard feature library be... , where n is the dimension of the feature vector;

[0083] First similarity It can be calculated and The cosine similarity between them is obtained using the following formula:

[0084]

[0085] In this formula:

[0086] Represents the dot product of two vectors;

[0087] and Let represent the Euclidean norm (i.e., the magnitude) of the two vectors respectively.

[0088] The calculation result is a scalar value in the interval [-1, 1]. In the context of this application, since the values ​​of each dimension of the feature vector are usually non-negative, The actual value range of is generally [0,1]. The closer the value is to 1, the more consistent the directions of the two vectors are, that is, the more similar the appearance of the requisitioned aviation materials is to the standard template; the closer the value is to 0, the greater the difference is.

[0089] The first preset condition and the determination of the issuance verification pass means that a similarity threshold or matching rule is preset. When the calculated first similarity meets the condition, it is determined that the electronic identity code and visual features are successfully matched. Specifically, the first similarity can be compared with the preset threshold. If it is greater than or equal to the threshold, an approval signal is triggered to automatically decide whether to allow the aircraft material to leave the warehouse.

[0090] Associating the unique part number with the current requisition task means that after verification, a logical binding relationship is established in the system background data between the unique part number and the requisition document, operator, timestamp, and target aircraft / maintenance task. Specifically, this can be achieved by writing an association record in a database transaction, which is used to trace the flow path and purpose of the aircraft material in subsequent processes.

[0091] In the assembly process, obtaining the target installation location identifier contained in the current electronic maintenance work order refers to retrieving the currently executing electronic work order data from the aircraft maintenance management system, parsing and extracting the specified component installation location code or description. This can be achieved by calling the system API or parsing the structured work order file, and is used to determine where the aircraft parts to be installed should be installed.

[0092] The assembly visual image of the aircraft material to be assembled captured by the second image acquisition device refers to the real-time image of the aircraft material to be installed that is captured by the deployed image acquisition equipment at the assembly station in the maintenance site (such as hangar or line maintenance point) before assembly. Specifically, it can be achieved by using an industrial camera with image stabilization and wide dynamic range, which is used to perform secondary visual confirmation of the aircraft material identity before final installation.

[0093] The calculation of the second similarity between the visual feature vector of the assembly visual image and the standard visual feature template corresponding to the associated unique part number code refers to repeating the feature extraction and similarity calculation process. However, the standard visual feature template used in this calculation is retrieved based on the unique part number code that has been associated with the current task in the previous requisition process. The specific process is similar to the requisition process but the data source is associated. This is used to verify whether the identity of the aircraft materials remains consistent and has not been changed during the process from the warehouse to the workstation.

[0094] In one specific embodiment of the present invention, the mathematical principle of calculating the second similarity is consistent with that of calculating the first similarity, both of which are achieved by measuring the directional consistency between the feature vector of the real-time image and the feature vector of the corresponding standard template in the high-dimensional feature space.

[0095] Specifically, the calculation process can be expressed by the following mathematical formula:

[0096] Let the extracted visual feature vector of the assembly visual image be... Let the feature vector of the standard visual feature template retrieved from the standard feature library based on the associated unique part number code be... Where n is the dimension of the feature vector, here... The template retrieved when calculating the first similarity is the same vector.

[0097] Second similarity Through calculation and The cosine similarity between them is obtained using the following formula:

[0098]

[0099] In this formula:

[0100] Representing vectors and The dot product;

[0101] and Let represent the Euclidean norms of the two vectors respectively.

[0102] The calculation result is also a scalar value in the range [0,1]. The closer the value is to 1, the more consistent the appearance of the aircraft material captured in the assembly stage is with the standard template associated with and confirmed in the previous requisition stage, verifying that the identity of the aircraft material has not changed during the circulation process.

[0103] The second preset condition and output assembly permission instruction refer to a preset similarity threshold (which may be the same as or different from the first preset condition). When the second similarity meets this condition, it is determined that the identity verification before assembly is passed. Specifically, the system can trigger the system to send an instruction to the operator's terminal to "allow installation" or light up a green indicator light, which serves as the final system authorization for carrying out the physical installation operation.

[0104] The core innovation of this solution lies in constructing an intelligent error prevention logic based on dual binding of "electronic identity and visual features" and closed-loop verification across "requisition-assembly" stages. By forcing the initial binding and verification at the requisition stage, and performing secondary verification and location compliance verification based on the task context at the assembly stage, it solves the inherent defects of traditional methods such as isolated data sources, disconnected stages, and reliance on single-point verification, thus achieving automated and highly reliable error prevention throughout the entire process of aircraft material circulation.

[0105] The working process and principle of this scheme are as follows: First, during the requisition process, the unique part number code of the aircraft material read by the RFID reader and the requisition visual image captured by the first image acquisition device are acquired simultaneously. Next, based on the unique part number code, the corresponding standard visual feature template is retrieved from the pre-built standard feature library, and the visual feature vector of the requisition visual image is extracted, and the first similarity between the two is calculated. When the first similarity meets the first preset condition, the requisition verification is determined to be successful, and the unique part number code is associated with the current requisition task in the system. Subsequently, during the assembly process, the target installation location identifier in the current electronic maintenance work order is acquired, and the assembly visual image of the aircraft material to be assembled is captured by the second image acquisition device. Then, the visual feature vector of the assembly visual image is extracted, and the second similarity between it and the standard visual feature template corresponding to the previously associated unique part number code is calculated. When the second similarity meets the second preset condition, the system outputs an assembly permission instruction. This continuous process ensures that the electronic identity and physical entity of aircraft parts are always consistent from the time they are taken out of the warehouse to the time they are assembled on site, and that the flow process is traceable. This proactively prevents human error in key links and improves maintenance safety and quality.

[0106] For example, the construction steps of a pre-built standard feature library specifically include:

[0107] Acquire multi-angle sample images of various aviation materials under standard conditions;

[0108] For various types of aviation materials, a feature extraction network is used to process multi-angle sample images corresponding to various types of aviation materials to generate a standard visual feature vector set for the various types of aviation materials.

[0109] The standard visual feature vector set is associated with and stored with the corresponding unique part number code to form a standard visual feature template.

[0110] Specifically, this application further proposes to construct a high-fidelity standard visual feature library for aircraft materials based on systematic data collection and deep feature learning, so as to ensure that the visual feature templates can comprehensively and stably represent the essential appearance attributes of various aircraft materials under standard conditions, and provide a robust benchmark for subsequent real-time comparison.

[0111] Among them, acquiring multi-angle sample images of various types of aviation materials under standard conditions refers to acquiring a series of images covering the main identification surfaces and key details of aviation materials under controlled lighting and background conditions when new parts are put into storage or aviation materials are restored to standard specifications after refurbishment. Specifically, high-resolution industrial cameras can be used in conjunction with uniform light sources to achieve this. The aim is to build an original image dataset that is unobstructed, undistorted, and has consistent lighting, as a reliable input source for feature learning.

[0112] Specifically, for various types of aircraft materials, a feature extraction network is used to process multi-angle sample images corresponding to each type of aircraft material, generating a standard visual feature vector set for each type of aircraft material. This involves inputting the collected multi-angle sample images into a deep convolutional neural network pre-trained on large-scale general image data, specifically using architectures such as ResNet or VisionTransformer, to extract high-dimensional feature vectors from the deep layers of the network. By aggregating and analyzing the feature vectors of all angle images of the same aircraft material, such as calculating the mean vector or constructing a feature vector distribution model, a standardized feature representation that can comprehensively reflect the multi-view appearance information of the aircraft material is formed.

[0113] The process of associating and storing the standard visual feature vector set with the corresponding unique part number code to form a standard visual feature template refers to establishing and maintaining a structured index in the database after feature calculation. This index strongly associates the feature vector set representing the visual identity of the aircraft material with the unique and authoritative part number code of the aircraft material in the information system. Specifically, a NoSQL database or a vector database can be used for storage and indexing to ensure that the corresponding visual feature template can be retrieved in milliseconds using the part number code.

[0114] Specifically, this solution constructs a standardized pipeline from physical entities to digital features. At the data acquisition end, a controlled environment and multi-angle imaging scheme ensure the quality and coverage of sample images from the source, avoiding lighting noise and incomplete perspective issues caused by random shooting. At the feature extraction end, a deep neural network automatically learns and extracts the deep visual features most distinctive for component identification, with representational capabilities far exceeding traditional manual features. At the data management end, by establishing a precise mapping relationship between "part number code - feature vector set," the material management attributes of aviation materials are deeply integrated with their visual biometrics, forming a searchable and scalable digital asset library. These three parts work together to ensure that the final generated standard visual feature template not only possesses high discriminative power, effectively identifying different part numbers, but also has good tolerance for subtle appearance differences in the same part number caused by manufacturing processes.

[0115] Compared to existing technologies, traditional methods for constructing standard feature libraries often rely on manually selecting representative images from a single perspective or using simple image processing algorithms (such as edge detection) to extract features. This results in weak template feature representation capabilities, poor generalization, and susceptibility to shooting conditions. In contrast, this solution ensures the completeness of visual information through systematic multi-angle data acquisition; it obtains more robust and essential feature representations by employing a deep feature extraction network; and it achieves information dimension fusion and efficient management by establishing a strong association with authoritative part number coding. This allows the constructed standard feature templates to maintain stable comparison performance even in complex real-world scenarios (such as changes in lighting, normal wear and tear on components, and partial occlusion).

[0116] Through the above technical solution, this application effectively solves the problems of traditional aircraft material visual templates being easily limited by acquisition conditions and lacking sufficient feature representation capabilities. Systematic multi-angle acquisition ensures the integrity of visual information, deep feature extraction networks guarantee the robustness and discriminative power of features, and strong correlation with the material management system achieves unified and efficient utilization of data dimensions. The standard feature library constructed by this solution elevates the accuracy and reliability of aircraft material appearance matching to a new level, laying a solid data foundation for subsequent high-precision, automated aircraft material identity and status verification.

[0117] For example, extracting the visual feature vector of the requisition visual image specifically includes:

[0118] The requisition visual image is input into a pre-trained feature extraction network;

[0119] The feature vector output by the feature extraction network at a specific layer is obtained and used as the visual feature vector of the visual image being used.

[0120] The feature extraction network has the same network structure as the network used to construct the standard visual feature template.

[0121] Specifically, this application further proposes to adopt a unified and fixed deep feature extraction network architecture to perform high-dimensional feature encoding on the real-time acquired visual images of aircraft materials, ensuring that the visual feature vectors extracted at different stages and at different time points have a consistent mathematical space and comparability, providing a stable and reliable measurement basis for subsequent similarity calculations.

[0122] The process of inputting the visual image into a pre-trained feature extraction network refers to feeding the real-time RGB or grayscale image captured by the first image acquisition device into a pre-trained deep neural network model after fixed-size normalization and standardization preprocessing. The aim is to transform pixel-level image information into structured, high-level semantic features.

[0123] In this context, obtaining the feature vector output by the feature extraction network at a specific layer means not using the network's final classification output, but extracting the high-dimensional tensor output by a certain intermediate layer (such as the last convolutional layer or after the global pooling layer), and flattening or aggregating it into a fixed-dimensional feature vector. Specifically, the activation value of this layer can be obtained by calling the forward propagation interface of the deep learning framework. This vector carries the key visual pattern information of the image, while discarding details that are irrelevant to identity recognition.

[0124] The feature extraction network uses the same network structure as the network used to construct the standard visual feature template. This refers to the network used for real-time processing, whose number of layers, number of neurons per layer, connection methods, and non-linear activation functions are completely consistent with the network used to process multi-angle sample images when constructing the standard feature library. More importantly, the weight parameters of all network layers are locked after deployment, maintaining absolute consistency with the final weights used in the standard feature library construction phase. This consistency is ensured by sharing the same model file or verifying the model hash value, thereby guaranteeing that the definition of the feature space does not drift.

[0125] Specifically, this solution constructs a consistent feature extraction pipeline that spans both data preparation and real-time application stages. At the real-time processing end, each frame of the requisition visual image undergoes the exact same network transformation as the standard sample image. This network, acting as a deterministic mathematical function, maps any input image to the same high-dimensional feature space. Due to the absolute uniformity of the network structure and weights, for the same aircraft material, whether in the standard sample image or the actual requisition capture image, as long as the appearance is essentially the same, the extracted feature vectors should be very close in position within this space; while the feature vectors of different aircraft materials will be spaced further apart. This design provides the mathematical premise for a direct and fair comparison between the "instantaneous features" extracted from real-time images and the "benchmark features" pre-stored in the standard feature library.

[0126] Compared to existing technologies, traditional methods may exhibit inconsistencies in feature extraction. For example, different algorithms may be used to process images at different stages, or updates to the same algorithm version may alter the feature space definition. This inconsistency introduces systematic errors that are difficult to quantify, directly compromising the reliability of the alignment results. Our proposed solution, however, fundamentally eliminates performance fluctuations and mismatch risks caused by differences in feature extractors by mandating and locking the use of the same fixed network throughout the entire process, from standard library construction to real-time alignment.

[0127] Through the above technical solution, this application effectively solves the problem of comparison benchmark failure caused by inconsistent feature representations in aircraft material visual recognition. A fixed and unified deep feature extraction network ensures the stability and comparability of the feature space, giving the similarity calculation between real-time image features and standard template features a solid mathematical foundation and physical meaning. This scheme elevates feature extraction from a variable processing step to a reliable metric, significantly enhancing the robustness and accuracy of the entire aircraft material identity verification system.

[0128] For example, when the first similarity meets the first preset condition, the requisition verification is determined to be successful, including:

[0129] Compare the first similarity with the first preset threshold;

[0130] If the first similarity is greater than or equal to the first preset threshold, it is determined that the first preset condition is met, and a first consistency identifier is generated;

[0131] If the first similarity is less than the first preset threshold, it is determined that the first preset condition is not met, a first alarm command is generated, and the requisition process is interrupted.

[0132] Specifically, this application further proposes an automated decision-making and process blocking mechanism based on a preset quantitative threshold, which makes a real-time and objective judgment on the calculated visual feature similarity to ensure that only aircraft materials with electronic identity and visual features that are highly matched can enter the subsequent circulation process, thereby establishing a reliable technical error prevention checkpoint at the source of requisition.

[0133] The comparison of the first similarity with the first preset threshold refers to comparing the numerical result representing the degree of matching of the appearance of the aircraft materials with a threshold value set in advance through statistical analysis or experimental verification. Specifically, it can be implemented by performing floating-point comparison operations in memory, which is used to transform the continuous similarity measurement into a discrete binary classification decision problem.

[0134] If the first similarity is greater than or equal to the first preset threshold, it is determined that the first preset condition is met and a first consistency identifier is generated. This means that when the comparison result reaches or exceeds the standard, the system automatically generates a logical signal or data tag representing "verification passed". Specifically, this can be achieved by setting a bit in the system status register or inserting a successful verification record in the business database. This identifier will serve as a key credential for allowing subsequent operations.

[0135] Wherein, if the first similarity is less than the first preset threshold, it is determined that the first preset condition is not met, and a first alarm instruction is generated and the requisition process is interrupted. This means that when the comparison result does not meet the standard, the system immediately triggers a predefined exception handling procedure, which may include sending an audible and visual alarm to the on-site operator terminal, popping up a warning box on the management interface, and simultaneously calling the process control interface to lock the status of the current requisition transaction to "paused" or "failed" in order to forcibly prevent the aircraft material from being physically taken away and the document from being confirmed.

[0136] Specifically, this solution constructs a closed-loop control logic of "calculation-judgment-execution". The output of the feature similarity calculation module serves as the input of the judgment module. The judgment module compares this input with a fixed threshold standard and, based on the comparison result, selectively triggers two distinct output paths within milliseconds: one generates a pass identifier and opens the process, and the other generates an alarm command and closes the loop. In this mechanism, the preset threshold serves as an objective benchmark for decision-making, the consistency identifier or alarm command serves as the execution command of the control system, and the process interruption function serves as the final physical safety guarantee. The three work together to ensure that the entire requisition verification process no longer relies on the operator's subjective judgment, but is instead guarded by a fast, accurate, and ruthless automated system.

[0137] Compared to existing technologies, traditional methods either lack automated judgment in the requisition and verification process, relying on manual visual inspection; or, while offering simple prompts, lack the ability to enforce process interruption, allowing errors to be overlooked or covered up. These methods are slow to respond, inconsistent, and incomplete in preventing errors. This solution, however, introduces automated judgment based on clear thresholds and enforced process control, enabling immediate detection and physical intervention of errors. It upgrades error prevention from "suggestive prompts" to "mandatory control," fundamentally eliminating the possibility of substandard aircraft materials leaking due to human negligence.

[0138] Through the above technical solution, this application effectively solves the problems of delays, omissions, and inconsistent standards caused by relying on manual decision-making in the aircraft material requisition process. The automated threshold determination mechanism ensures the immediacy and objectivity of decision-making, while the deep linkage between verification results and business processes (generating identifiers or triggering alarm interruptions) ensures the effective implementation of error prevention measures. This solution establishes a solid technical barrier at the initial requisition stage, significantly improving the accuracy and security of aircraft material distribution and laying the foundation for the reliability of subsequent processes.

[0139] For example, it also includes:

[0140] During the assembly process, an image of the on-site environment at the target installation location is acquired by a third image acquisition device;

[0141] Based on the target installation location identifier, retrieve the corresponding standard installation location feature template from the standard feature library;

[0142] Extract the visual feature vector of the on-site environment image, and calculate the third similarity between the visual feature vector of the on-site environment image and the standard installation position feature template;

[0143] The output assembly license instruction must be executed when the third similarity meets the third preset condition.

[0144] Specifically, this application further proposes to introduce environmental visual verification of the target installation position during the assembly process, and to construct a dual matching decision logic of "component identity-installation position", thereby adding a key error prevention verification before the final physical assembly, ensuring that the aircraft material is installed in the absolutely correct position, and eliminating human error of "installing in the wrong place" from the root.

[0145] Among them, acquiring on-site environmental images of the target installation location captured by the third image acquisition device refers to acquiring images of the target installation interface or compartment on the aircraft body through a dedicated imaging device at the maintenance station before the assembly action is performed. Specifically, explosion-proof industrial cameras or endoscope cameras with supplementary LED lighting can be used to acquire visual context information such as the physical environment around the installation point, interface shape, and adjacent components.

[0146] Specifically, retrieving the corresponding standard installation location feature template from the standard feature library based on the target installation location identifier means using the standardized location code parsed from the electronic maintenance work order as an index to retrieve the standard environmental feature data that has been learned and stored for that location from the feature library. This can be achieved by querying a relational database. The template defines "what the correct location should look like".

[0147] The process of extracting the visual feature vector of the on-site environment image and calculating the third similarity with the standard installation location feature template refers to using a deep network model that is the same as or compatible with the component feature extraction to encode the features of the on-site environment image, and then calculating the distance or similarity between it and the standard location template features in the same high-dimensional space. Specifically, cosine similarity can be used to quantify the degree of matching between the current actual installation environment and the standard environment model.

[0148] In one specific embodiment of the present invention, the calculation of the third similarity adopts the same consistency measurement principle as the first and second similarities, which is achieved by measuring the degree of directional matching between the visual feature vector of the on-site environment image and the standard installation position feature template in the high-dimensional feature space.

[0149] Specifically, the calculation process can be expressed by the following mathematical formula:

[0150] Let the extracted visual feature vector of the scene environment image be... Let the feature vector of the standard installation location feature template retrieved from the standard feature library based on the target installation location identifier be . , where n is the dimension of the feature vector.

[0151] The third similarity Through calculation and The cosine similarity between them is obtained using the following formula:

[0152]

[0153] In this formula:

[0154] Representing vectors and The dot product;

[0155] and Let represent the Euclidean norms of the two vectors respectively;

[0156] The calculation result is a scalar value within the range [0,1]. The closer the value is to 1, the higher the visual characteristics of the current installation location are compared with the pre-stored standard installation location, meaning the higher the probability of "correct location".

[0157] The output assembly permission instruction, which is to be executed when the third similarity meets the third preset condition, means that the system regards the position matching degree as a necessary condition on par with the component identity matching degree. Only when both meet their respective preset threshold standards will the assembly be comprehensively judged and authorized. Specifically, this can be achieved through an AND logic gate circuit or a conditional judgment statement in the software.

[0158] Specifically, this solution constructs a parallel dual-path verification channel. One channel (component identity verification) ensures that "the component in hand is correct," while the other, newly added channel (installation location verification), ensures that "the installation location is also correct." Both channels share the same feature processing and decision-making framework, but process different data objects and use different standard templates. At the decision convergence point, the system requires that both independent similarity calculation results simultaneously meet the standard. This design means that even if component identity verification passes (the correct part was obtained), if the environmental features of the installation location captured by the camera do not match the location template required by the work order (e.g., the lens is pointed at a nearby similar interface), the system will still refuse authorization due to insufficient third similarity, thus triggering an alarm. This adds an independent dimension of evidence based on environmental context, greatly improving the reliability and security of the decision.

[0159] Compared to existing technologies, traditional assembly verification typically only checks the barcodes on the components themselves or provides simple location information, lacking automated and objective visual verification of the actual physical installation environment. This method cannot detect errors such as operators misreading work orders or misidentifying installation points. This solution introduces and enforces vision-based location-environment matching verification, extending the error-proofing scope from the components themselves to the correct relationship between the components and the environment, thus solving the verification blind spots in the final assembly stage of traditional methods.

[0160] Through the above technical solution, this application effectively addresses the deep-seated problem of the lack of automated verification of target installation positions in the aircraft material assembly process. The combination of independent visual verification of the location environment and component identification verification constitutes a multi-dimensional and complementary error-proofing system. This solution elevates the reliability of assembly decisions from a single component dimension to a "component-location" correlation dimension, ensuring that any single or double error cannot pass system verification. This provides the highest level of safety assurance before the final physical connection occurs, significantly reducing maintenance quality risks and safety hazards caused by assembly errors.

[0161] For example, the output assembly license instruction specifically includes:

[0162] When the second similarity satisfies the second preset condition and the third similarity satisfies the third preset condition, a second consistency identifier is generated;

[0163] The second consistency identifier, the unique part number code, and the target installation location identifier are associated and recorded in the electronic maintenance record.

[0164] Specifically, this application further proposes a final authorization decision and digital evidence storage mechanism based on the fusion of dual-path verification results. This ensures that an assembly permit with legal and technical validity is generated only when both the component's identity and installation location have been strictly verified. The complete verification evidence chain is automatically archived to form an immutable maintenance quality traceability record.

[0165] Specifically, when the second similarity satisfies the second preset condition and the third similarity satisfies the third preset condition, the generation of the second consistency identifier means that the second similarity, which represents the matching degree of component identity, and the third similarity, which represents the matching degree of location environment, are compared with their respective independent preset thresholds. Only when both comparison results are "true" will the system comprehensively determine that the assembly verification has passed as a whole, and generate a unique digital token or logical flag representing this authorization event. Specifically, this can be achieved by setting a specific status bit in memory or generating an encrypted digital signature with a timestamp.

[0166] The process of associating the second consistency identifier, the unique part number code, and the target installation location identifier with the electronic maintenance record means that while generating the consistency identifier, the system automatically initiates a structured data write transaction to forcibly associate and bind the core metadata of this authorization—namely, the identifier representing the authorization itself, the identity code of the authorized installed component, and the specific location code of the authorization—and persistently store it as a complete record in the database of the maintenance management system. Specifically, database transactions can be used to ensure the atomicity and consistency of the record.

[0167] Specifically, this solution constructs a continuous automated chain of "determination-authorization-proofing". The decision module receives Boolean results from two independent verification channels and derives the final authorization conclusion through a logical AND operation. Once the conclusion is "passed", the system does not simply output an "allow" signal, but simultaneously triggers two key actions: first, it generates a unique authorization credential (second consistency identifier) ​​as instantaneous proof of the legality of this operation; second, it immediately packages this credential with the two most critical entity identifiers in the business and writes it into the electronic maintenance record as official evidence. This means that the authorization action and the recording action are atomic and inseparable. The consistency identifier is the index of the record, and the record gives the consistency identifier specific business meaning and historical context.

[0168] Compared to existing technologies, traditional assembly completion records often rely on manual post-assembly entry or simple barcode scanning for confirmation, which can lead to incomplete records, input errors, or even forgery, resulting in a disconnect between authorization and recording. This solution, however, designs the final authorization instruction output as a mandatory, self-documenting process. It ensures that every system-authorized assembly operation is automatically accompanied by the creation of a complete digital record containing "who (part ID), where (location ID), when (implied in the conformity identifier), and whether it has been verified by the system (the conformity identifier itself)." This transforms the compliance and traceability of maintenance work from relying on human discretion to being enforced by system processes.

[0169] Through the above technical solution, this application effectively solves the problems of traceability difficulties and unclear responsibilities caused by the separation of authorization and recording in the assembly process. The fusion decision-making mechanism based on dual-path verification ensures the rigor of authorization, while the atomic association between authorization and evidence storage achieves the automation, integrity, and non-repudiation of operation records. This solution transforms each successful assembly into auditable digital evidence, not only preventing errors in real time but also providing a solid data foundation for subsequent quality analysis, safety audits, and continuous improvement, thus achieving a closed loop of error prevention management before, during, and after the process.

[0170] For example, when the requisition verification is deemed to have failed, the following steps are also included:

[0171] Obtain one or more candidate part number codes that are visually similar to the currently read unique part number code;

[0172] Obtain the candidate standard visual feature template corresponding to the candidate part number code from the standard feature library;

[0173] Calculate the candidate similarity between the visual feature vector of the requisition visual image and each of the candidate standard visual feature templates;

[0174] The output includes a prompt message containing the candidate part number code and its corresponding candidate similarity.

[0175] Specifically, this application further proposes an intelligent candidate part number assisted identification mechanism when the requisition verification fails. By performing reverse retrieval and similarity ranking on the standard feature library, it provides the operator with a list of the most likely correct alternative part numbers, thereby upgrading a simple "error report" into an efficient "error correction assistance", which significantly improves the processing efficiency and user experience in abnormal situations.

[0176] The process of obtaining one or more candidate part numbers that are visually similar to the currently read unique part number code refers to quickly finding other part numbers that are most similar to the currently failed aircraft material in terms of historical image features in the standard feature library based on a pre-established visual feature space nearest neighbor index. Specifically, this can be achieved by running an approximate nearest neighbor search algorithm or querying a pre-built similarity map.

[0177] The step of obtaining the candidate standard visual feature template corresponding to the candidate part number code from the standard feature library means retrieving the standard visual feature template corresponding to each candidate part number in parallel from the feature library based on the retrieved list of candidate part numbers, in order to prepare for subsequent refined comparison.

[0178] The calculation of the candidate similarity between the visual feature vector of the requisition visual image and each of the candidate standard visual feature templates refers to calculating the similarity between the feature vector of the currently collected requisition visual image that caused the verification failure and the feature vector of each candidate template, thereby generating a quantified matching score for each candidate item number.

[0179] The output includes prompts containing the candidate part number codes and their corresponding candidate similarity scores. This means that the sorted list of candidate part numbers and their matching scores are pushed to the interactive interface of the material requisition terminal in real time in the form of a visual list or voice prompts. Specifically, the numbers can be sorted from high to low similarity and the most likely correct option is highlighted.

[0180] Specifically, this solution constructs a parallel, service-oriented intelligent error correction auxiliary channel outside the core verification process. When the main process fails due to insufficient similarity, the system does not stagnate in the error reporting state but immediately activates this auxiliary channel. This channel utilizes the pre-built global visual feature space, using the feature vector of the currently unrecognizable problem image as a query point to quickly locate the nearest known points in the space. Subsequently, by recalculating the precise distances to these neighboring points, a candidate list sorted by probability is generated. Finally, this list is intuitively fed back to the operator as decision support information. In this way, a collaboration is formed between the system and the operator: the system is responsible for massive memory and rapid screening, while the operator is responsible for final confirmation. The combination of the two greatly shortens the time required to check for errors and find the correct part number.

[0181] Compared with existing technologies, traditional systems usually only give a general alarm such as "verification failed" or "part number mismatch" when verification fails, throwing the entire troubleshooting burden to the operator, who needs to rely on memory or manually look up a huge catalog, which is inefficient and prone to omissions.

[0182] Through the above technical solution, this application effectively solves the problem of operators lacking effective technical means to quickly locate the correct item number after a failed item requisition verification. The intelligent candidate suggestion mechanism fully utilizes the knowledge of the built feature library, transforming a failed matching attempt into a successful similar item retrieval. This not only significantly reduces the downtime caused by manual searching and verification, but also fundamentally reduces the risk of operators selecting the wrong item again in their anxiety, achieving an upgrade from error prevention to error assistance, and improving the intelligence level and operational resilience of the entire requisition process.

[0183] For example, it also includes historical data analysis steps:

[0184] The failure records and final fault confirmation records of each unique part number code corresponding to the aircraft material during the requisition verification and assembly verification stages are statistically analyzed.

[0185] Based on the statistical results, the matching threshold of the standard visual feature templates for the corresponding aircraft materials is adjusted.

[0186] Example 2

[0187] This application also discloses an artificial intelligence-based aviation material management system.

[0188] Reference Figure 2 An artificial intelligence-based aircraft material management system includes:

[0189] The first data acquisition module is used to acquire the unique part number code stored in the electronic tag of the aviation material to be requisitioned, which is read by the radio frequency identification reader, and the requisition visual image of the aviation material to be requisitioned, which is captured by the first image acquisition device.

[0190] The retrieval module is used to retrieve the corresponding standard visual feature template from the pre-built standard feature library based on the unique part number code;

[0191] The first similarity calculation module is used to extract the visual feature vector of the requisition visual image and calculate the first similarity between the visual feature vector and the standard visual feature template.

[0192] The requisition verification and determination module is used to determine that the requisition verification is passed when the first similarity meets the first preset condition, and to associate the unique item number code with the current requisition task;

[0193] The second data acquisition module is used to acquire, during the assembly process, the target installation location identifier contained in the current electronic maintenance work order, as well as the assembly visual image of the aircraft material to be assembled captured by the second image acquisition device.

[0194] The second similarity calculation module is used to extract the visual feature vector of the assembly visual image and calculate the second similarity between the visual feature vector of the assembly visual image and the standard visual feature template corresponding to the associated unique part number code.

[0195] The instruction output module is used to output an assembly permission instruction when the second similarity meets the second preset condition.

[0196] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention, they should all fall within the protection scope of the present invention.

[0197] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0198] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. An artificial intelligence-based method for managing aircraft materials, characterized in that, Includes the following steps: The unique part number code stored in the electronic tag of the aviation material to be requisitioned is obtained by the radio frequency identification reader, and the visual image of the aviation material to be requisitioned is captured by the first image acquisition device. Based on the unique part number code, the corresponding standard visual feature template is retrieved from the pre-built standard feature library; Extract the visual feature vector of the requisition visual image, and calculate the first similarity between the visual feature vector and the standard visual feature template; When the first similarity meets the first preset condition, the requisition verification is determined to be successful, and the unique part number is associated with the current requisition task. When the requisition verification fails, the following also applies: Obtain one or more candidate part number codes that are visually similar to the currently read unique part number code; Obtain the candidate standard visual feature template corresponding to the candidate part number code from the standard feature library; Calculate the candidate similarity between the visual feature vector of the requisition visual image and each of the candidate standard visual feature templates; Output a prompt message containing the candidate part number code and its corresponding candidate similarity; During the assembly process, the target installation location identifier contained in the current electronic maintenance work order is obtained, as well as the assembly visual image of the aircraft material to be assembled captured by the second image acquisition device. Extract the visual feature vector of the assembly visual image, and calculate the second similarity between the visual feature vector of the assembly visual image and the standard visual feature template corresponding to the associated unique part number code; When the second similarity meets the second preset condition, an assembly permission command is output; Also includes: During the assembly process, an image of the on-site environment at the target installation location is acquired by a third image acquisition device; Based on the target installation location identifier, retrieve the corresponding standard installation location feature template from the standard feature library; Extract the visual feature vector of the on-site environment image, and calculate the third similarity between the visual feature vector of the on-site environment image and the standard installation position feature template; The output assembly permission instruction must be executed when the third similarity meets the third preset condition; It also includes historical data analysis steps: The failure records and final fault confirmation records of each unique part number code corresponding to the aircraft material during the requisition verification and assembly verification stages are statistically analyzed. Based on the statistical results, the matching threshold of the standard visual feature templates for the corresponding aircraft materials is adjusted.

2. The method for managing aircraft materials based on artificial intelligence according to claim 1, characterized in that, The steps for building a pre-built standard feature library specifically include: Acquire multi-angle sample images of various aviation materials under standard conditions; For various types of aviation materials, a feature extraction network is used to process multi-angle sample images corresponding to various types of aviation materials to generate a standard visual feature vector set for the various types of aviation materials. The standard visual feature vector set is associated with and stored with the corresponding unique part number code to form a standard visual feature template.

3. The method for managing aircraft materials based on artificial intelligence according to claim 2, characterized in that, Extracting the visual feature vector of the requisition visual image specifically includes: The requisition visual image is input into a pre-trained feature extraction network; The feature vector output by the feature extraction network at a specific layer is obtained and used as the visual feature vector of the visual image being used. The feature extraction network has the same network structure as the network used to construct the standard visual feature template.

4. The aircraft material management method based on artificial intelligence according to claim 1, characterized in that, When the first similarity meets the first preset condition, the requisition verification is deemed successful, including: Compare the first similarity with the first preset threshold; If the first similarity is greater than or equal to the first preset threshold, it is determined that the first preset condition is met, and a first consistency identifier is generated. If the first similarity is less than the first preset threshold, it is determined that the first preset condition is not met, a first alarm command is generated, and the requisition process is interrupted.

5. The aircraft material management method based on artificial intelligence according to claim 1, characterized in that, Output assembly permission instructions, specifically including: When the second similarity satisfies the second preset condition and the third similarity satisfies the third preset condition, a second consistency identifier is generated; The second consistency identifier, the unique part number code, and the target installation location identifier are associated and recorded in the electronic maintenance record.

6. An artificial intelligence-based aircraft material management system, applied to the artificial intelligence-based aircraft material management method described in any one of claims 1-5, characterized in that, include: The first data acquisition module is used to acquire the unique part number code stored in the electronic tag of the aviation material to be requisitioned, which is read by the radio frequency identification reader, and the requisition visual image of the aviation material to be requisitioned, which is captured by the first image acquisition device. The retrieval module is used to retrieve the corresponding standard visual feature template from the pre-built standard feature library based on the unique part number code; The first similarity calculation module is used to extract the visual feature vector of the requisition visual image and calculate the first similarity between the visual feature vector and the standard visual feature template. The requisition verification and determination module is used to determine that the requisition verification is passed when the first similarity meets the first preset condition, and to associate the unique item number code with the current requisition task; The second data acquisition module is used to acquire, during the assembly process, the target installation location identifier contained in the current electronic maintenance work order, as well as the assembly visual image of the aircraft material to be assembled captured by the second image acquisition device. The second similarity calculation module is used to extract the visual feature vector of the assembly visual image and calculate the second similarity between the visual feature vector of the assembly visual image and the standard visual feature template corresponding to the associated unique part number code. The instruction output module is used to output an assembly permission instruction when the second similarity meets the second preset condition.

7. A computer-readable storage medium, characterized in that: The system stores instructions that, when executed on a computer, cause the computer to perform an artificial intelligence-based aircraft material management method as described in any one of claims 1 to 5.

Images