Method, system, device and storage medium for detecting aircraft landing gear failure

Abstract

The application discloses an aircraft landing gear fault detection method, system, device and storage medium. The method comprises: acquiring multi-dimensional state data of the landing gear, including visual images, non-destructive testing, thermal testing and various physical parameter data; inputting the state data and a digital twin model based on the original design and continuously updated into a pre-trained fault classification model, outputting the type, position and severity information of the fault through cross-validation and analysis; evaluating the risk level based on the fault information and a fault analysis database; and generating a maintenance scheme in combination with an expert knowledge base, and the effect after maintenance can be simulated and verified. The application combines multi-source data fusion, dynamic digital twin and intelligent diagnosis model, realizes accurate and efficient identification and decision support for various typical faults such as fatigue cracks, seal wear and composite material delamination, and significantly improves the detection automation level and maintenance efficiency.

Description

Technical Field

[0001] This invention relates to the field of aviation equipment testing and maintenance technology, specifically to a method, system, equipment, and storage medium for detecting aircraft landing gear failures. Background Technology

[0002] Landing gear is a critical load-bearing component that contacts the ground, and its working condition directly affects the safety of takeoff and landing. Landing gear has a complex structure, integrating multiple subsystems such as mechanical, hydraulic, and electrical systems. During its service life, it is subjected to huge impact loads, alternating loads, and corrosive environments, making it highly susceptible to failures such as fatigue cracks, structural deformation, component wear, and hydraulic system leaks.

[0003] In the prior art, such as the patent application with application number CN202311520466.X, entitled "Civil Aviation Maintenance Inspection Method, Equipment and Medium Based on Target Detection Technology", it only identifies landing gear cracks through image data. However, due to the low accuracy of single data, it is easily affected by environmental factors and other factors, leading to misjudgment, thus resulting in the problem of low accuracy of fault detection. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, device, and storage medium for detecting aircraft landing gear faults. By acquiring multidimensional state data of the landing gear, and based on the state data and a digital twin model of the landing gear, the accuracy of fault detection is improved from two perspectives: the multidimensionality of the state data and the combined judgment based on the state data and the digital twin model. This solves the problem of low accuracy in the fault detection of existing aircraft landing gear.

[0005] This invention is achieved through the following technical solution:

[0006] The first aspect of this application provides a method for detecting aircraft landing gear malfunctions, including:

[0007] Acquire landing gear status data for landing gear fault diagnosis; the status data includes at least visual image data, non-destructive testing data, thermal testing data, and various physical parameter data of the landing gear;

[0008] The landing gear status data and a digital twin model matching the current physical state of the landing gear are input into a pre-trained fault classification model to obtain the landing gear fault information output by the fault classification model. The fault classification model is a fusion model that includes an improved U-Net for image segmentation, a long short-term memory network for temporal data analysis, and a convolutional-recurrent neural network for multimodal fusion. The digital twin model is a model that matches the physical entity of the landing gear, built based on the original design and manufacturing data of the landing gear and continuously updated based on historical and actual operating data. The fault information includes the fault type determined based on the status data, and the fault location and fault severity determined based on the status data and the digital twin model.

[0009] Based on the fault information output by the fault classification model and the preset fault analysis database, the risk level of the fault is assessed.

[0010] Based on the fault information, the assessed risk level, and the preset expert knowledge base, a repair plan for the fault is determined; the expert knowledge base stores repair standards and repair reference information.

[0011] In one feasible implementation, the landing gear state data and a digital twin model matching the current physical state of the landing gear are input into a pre-trained fault classification model, causing the fault classification model to perform:

[0012] Based on at least one type of data in the state data and reference values ​​of parameters in the digital twin model, the suspected fault type is determined; the fault type includes at least fatigue cracks, wear, and abnormal displacement.

[0013] Based on at least one of the other types of data in the state data, and the simulation results based on the digital twin model, the suspected fault is verified to determine the type of fault to be output.

[0014] In one feasible implementation, the fault type includes fatigue cracks, then,

[0015] The landing gear status data, along with a digital twin model matching the current physical state of the landing gear, are input into a pre-trained fault classification model, causing the fault classification model to perform:

[0016] For the ultrasonic image data in the non-destructive testing data, defect segmentation is performed using an improved U-Net to identify crack features;

[0017] The crack features are correlated and compared with the structural parameters at the corresponding crack location in the digital twin model to obtain the crack location and crack severity, thereby identifying suspected fatigue crack faults.

[0018] For the vibration data corresponding to the crack location in the physical parameter data, vibration anomalies are identified by long short-term memory network, and combined with the dynamic simulation results based on digital twin model, fatigue crack failure of landing gear is determined.

[0019] In one feasible implementation, the fault type includes wear of the actuator piston rod seal; then,

[0020] The landing gear status data, along with a digital twin model matching the current physical state of the landing gear, are input into a pre-trained fault classification model, causing the fault classification model to perform:

[0021] For the infrared thermogram data in the thermal detection data, temperature anomalies are identified by the improved U-Net, and the temperature anomaly areas are determined by combining the standard values ​​of the corresponding area temperature marked in the digital twin model; the infrared thermogram data characterizes the temperature changes during the landing gear retraction and extension process.

[0022] Within the same time period, for the displacement data representing the extension and retraction of the actuator cylinder in the physical parameter data of the temperature anomaly area, a long short-term memory network is used to identify suspected wear faults of the actuator cylinder piston rod seal.

[0023] For the pressure data of the hydraulic system in the physical parameter data, pressure anomalies are identified by long short-term memory network, and the wear failure of the piston rod seal of the actuator is determined by combining the hydraulic system flow field simulation results based on the digital twin model.

[0024] In one feasible implementation, the fault type includes composite material delamination defects, then,

[0025] The landing gear status data, along with a digital twin model matching the current physical state of the landing gear, are input into a pre-trained fault classification model, causing the fault classification model to perform:

[0026] For the terahertz imaging data in the non-destructive testing data, a convolutional neural network is used to identify the abnormal material layering regions in the landing gear doors.

[0027] Based on the terahertz wave propagation parameters in the digital twin model, the delamination area is calculated to identify suspected delamination defects in composite materials.

[0028] For the door locking pressure data in the physical parameter data, a recurrent neural network is used to identify pressure anomalies in order to determine the delamination defects of the composite material.

[0029] In one feasible implementation, the risk level of the fault is determined by the product of the probability of the fault occurring, the severity of the fault's consequences, and the difficulty of fault detection.

[0030] In one feasible implementation, the method further includes: simulating the expected effect after performing maintenance based on the maintenance plan to verify the effectiveness of the maintenance plan; the expected effect is characterized by the performance recovery rate of the digital twin model, and the maintenance is deemed qualified when the performance recovery rate exceeds a set threshold.

[0031] A second aspect of this application provides a detection system for aircraft landing gear malfunctions, comprising:

[0032] The data acquisition unit is used to acquire landing gear status data for landing gear fault diagnosis; the status data includes at least visual image data of the landing gear, non-destructive testing data, thermal testing data, and various physical parameter data.

[0033] The fault determination unit is used to input the landing gear status data and a digital twin model matching the current physical state of the landing gear into a pre-trained fault classification model to obtain the landing gear fault information output by the fault classification model. The fault classification model is a fusion model that includes an improved U-Net for image segmentation, a long short-term memory network for temporal data analysis, and a convolutional-recurrent neural network for multimodal fusion. The digital twin model is a model that matches the physical entity of the landing gear, constructed based on the original design and manufacturing data of the landing gear and continuously updated based on historical and actual operating data. The fault information includes the fault type determined based on the status data, and the fault location and fault severity determined based on the status data and the digital twin model.

[0034] The fault risk assessment unit assesses the risk level of a fault based on the fault information output by the fault classification model and a preset fault analysis database.

[0035] The maintenance decision-making unit determines the maintenance plan for the fault based on the fault information, the assessed risk level, and a preset expert knowledge base; the expert knowledge base stores maintenance standards and maintenance reference information.

[0036] A third aspect of this application provides an electronic device, including: a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the above-described method.

[0037] A fourth aspect of this application provides a storage medium, comprising: storing a program or instructions on the storage medium, wherein the program or instructions, when executed by a processor, implement the steps of the above-described method.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] This application embodiment acquires multi-dimensional real-time status data of the landing gear using various high-precision sensors and imaging devices, and inputs this data along with a digital twin model into a fault classification model to obtain accurate landing gear fault information. Since the digital twin model is built based on the original design and manufacturing data of the landing gear and is continuously updated based on historical and operational data, the fault classification model, based on the analysis results of real-time data and the digital twin model matching the current physical state of the landing gear, can accurately identify faults. Furthermore, this embodiment automatically identifies potential faults such as wear, cracks, and abnormal displacement, assesses the risk level, and generates detailed inspection reports and maintenance plans. This achieves closed-loop management of the entire process from fault prevention and accurate inspection to efficient maintenance, significantly improving the safety, accuracy, and efficiency of aircraft landing gear maintenance and effectively reducing aviation operational risks. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0041] Figure 1 A flowchart illustrating a method for detecting aircraft landing gear failures provided in this application embodiment;

[0042] Figure 2 A schematic diagram illustrating the construction process of a digital twin model in an aircraft landing gear fault detection method provided in this application embodiment;

[0043] Figure 3 A flowchart illustrating the fault identification and risk level classification in an aircraft landing gear fault detection method provided in this application embodiment;

[0044] Figure 4 A schematic diagram illustrating the process of making maintenance decisions and handling faults in an aircraft landing gear fault detection method provided in this application embodiment;

[0045] Figure 5 A schematic diagram of the landing gear strut crack identification process in a specific implementation method 1 provided in this application embodiment;

[0046] Figure 6 A schematic diagram of ultrasonic phased array detection of landing gear strut cracks in a specific implementation 1 provided in this application embodiment;

[0047] Figure 7An infrared thermal imaging diagram of landing gear temperature anomaly detection in a specific implementation 2 provided in this application embodiment;

[0048] Figure 8 A schematic diagram of the temperature timing curve of the nose landing gear actuator cylinder in a specific implementation method 2 provided in this application embodiment;

[0049] Figure 9 A terahertz imaging schematic diagram of composite material door layering in a specific implementation 3 provided in this application embodiment;

[0050] Figure 10 A schematic diagram of the structure of an aircraft landing gear fault detection system provided in an embodiment of this application;

[0051] Figure 11 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are for explanation only and are not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments in this application without creative effort are within the scope of protection of this application.

[0053] As will be known to those skilled in the art, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0054] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to such process, method, product, or apparatus.

[0055] Example 1:

[0056] Embodiment 1 of this application provides a method for detecting aircraft landing gear failures, in order to solve the problem of low accuracy in existing aircraft landing gear failure detection.

[0057] The subject executing this method can be any computing device capable of implementing the method, such as a server, mobile phone, personal computer, smart wearable device, smart robot, etc.

[0058] Furthermore, the embodiments of this application do not limit the execution order of different steps. When using the method provided in the embodiments of this application, the execution order of different steps can be adjusted according to actual needs.

[0059] For ease of description, the following uses an aircraft landing gear fault detection device as the subject of this method to provide a detailed description of the method provided in this application embodiment.

[0060] like Figure 1 The diagram shown is a flowchart illustrating the specific implementation of a method for detecting aircraft landing gear failures according to an embodiment of this application, including the following steps 11-14:

[0061] Step 11: Obtain landing gear status data for landing gear fault diagnosis.

[0062] In this embodiment, based on a variety of high-precision sensors and imaging devices, comprehensive status data of the landing gear is collected during each landing gear maintenance. A high-resolution camera captures images of the landing gear's exterior; a camera built into the landing gear bay takes deep images of the internal cavities; an ultrasonic phased array detector scans key load-bearing components (such as landing gear struts, rocker arms, and axles); an ultrasonic detector inspects welds and complex structural parts; a terahertz imager images composite material components; an infrared thermal imager records the temperature distribution of each component; and displacement, pressure, and vibration sensors simultaneously collect corresponding physical parameter data.

[0063] The status data includes at least visual image data of the landing gear, non-destructive testing data, thermal testing data, and various physical parameter data. In this embodiment, the specific acquisition of status data is shown in Table 1 below. Table 1 provides details of the acquisition of landing gear status data, including the module level to which the component acquiring the corresponding status data belongs, the name of the component acquiring the corresponding status data, the main function and detection object of the component, and the type of status data output by the component.

[0064] The core control unit is implemented through a central processing and control unit, responsible for system startup, task scheduling, synchronization control, data reception, temporary storage, and preliminary preprocessing. System startup can be the activation of the detection system in this embodiment, enabling each sensor and imaging device to turn on for data acquisition. Task scheduling can be the scheduling of sensors (or imaging devices) of the same type distributed in different locations. Synchronization control can be the control of the simultaneous opening and closing of multiple sensors. Data reception can be the reception of status data collected by each sensor and imaging device. Temporary storage can be the temporary storage of status data. Preliminary preprocessing can be the removal of obviously erroneous data from the collected status data.

[0065] The optical image acquisition unit acquires visual image data through a high-resolution camera and a camera built into the landing gear bay. Specifically, it is used to capture images of the landing gear's exterior, key connection parts, and internal cavities, as well as surface damage, deformation, and foreign objects.

[0066] The non-destructive testing unit collects non-destructive testing data through an ultrasonic phased array detector and a terahertz imager; the ultrasonic phased array detector is used to detect defects such as cracks and delamination inside the landing gear; and the terahertz imaging technology is highly sensitive to defects such as delamination and debonding of composite materials.

[0067] The thermal detection unit collects thermal detection data using an infrared thermal imager; by detecting the temperature distribution of various landing gear components, it identifies areas of abnormal heat generation.

[0068] The physical quantity sensor unit collects corresponding physical parameter data such as displacement, pressure and vibration through displacement sensors, pressure sensors and vibration sensors; the displacement sensors are installed on key moving parts to monitor the position accuracy and deformation during the retraction and extension process; the pressure sensors monitor the pressure changes of the hydraulic system; and the vibration sensors collect the vibration signals of the landing gear in the working state to analyze the structural health status.

[0069] Table 1. Details of landing gear status data collection.

[0070] In one feasible implementation, step 11 further includes: preprocessing the acquired multi-source state data, including image denoising, enhancement, and correction, as well as filtering and normalization of sensor data. Then, the preprocessed image data and 3D point cloud data are spatially registered and aligned with the digital twin reference model (described in detail in step 12) to ensure that the correspondence between the data and the model is accurate.

[0071] Step 12: Input the landing gear status data and the digital twin model matching the current physical state of the landing gear into the pre-trained fault classification model to obtain the landing gear fault information output by the fault classification model.

[0072] The digital twin model is a model that matches the physical entity of the landing gear, built based on the original design and manufacturing data of the landing gear and continuously updated based on historical and actual operation data; the fault information includes the fault type determined based on the status data, and the fault location and fault severity determined based on the status data and the digital twin model.

[0073] like Figure 2The diagram illustrates the construction process of the digital twin model. First, core raw data for constructing the landing gear digital twin model is acquired, including original mechanical CAD drawings, design parameters and manuals, material properties and performance parameters, and manufacturing process data. Based on this core raw data, an initial high-fidelity model is constructed: precise geometric dimensions are obtained from the original mechanical CAD drawings, design parameters, and manuals to build the geometric framework of the digital twin model. Then, material and physical properties, assembly relationships and kinematic constraints, and multiphysics information (mechanical / thermal / dynamic properties) are added to the corresponding parts of the geometric framework to determine the initial high-fidelity model (or digital twin baseline model).

[0074] This initial high-fidelity model can be used during initial maintenance as input to a pre-trained fault classification model to assist in fault identification. During the landing gear's service life, the various components may exhibit conditions different from the original data due to actual operating environments, service duration, component maintenance and replacement, etc. Therefore, to ensure that the digital twin model matches the current physical state of the landing gear, it is necessary to iterate and optimize based on historical and actual operating data to update the digital twin model.

[0075] The specific iteration and optimization process can be as follows: update the values ​​and status of model parameters based on historical and actual operating data of the landing gear; learn fault modes and expansion patterns based on historical fault judgments; and optimize detection thresholds and weights based on actual operating data to improve the accuracy of fault prediction based on the digital twin model.

[0076] In addition, the received real-time and historical data include: historical maintenance records, multi-dimensional detection data such as ultrasonic / image data, and actual operating data of the landing gear.

[0077] The fault classification model is a deep learning model trained on a dataset containing a large amount of labeled landing gear fault data (including images and sensor data of normal and various fault states). When used for fault detection, this model performs multi-dimensional comparative analysis between the acquired multi-dimensional state data and the digital twin model to determine the landing gear fault information.

[0078] The fault classification model can be a fusion model that includes an improved U-Net for image segmentation, a Long Short-Term Memory (LSTM) network for temporal data analysis, and a Convolutional-Recurrent Neural Network (CNN-RNN) for multimodal fusion.

[0079] Training the fault classification model involves training the improved U-Net network, the Long Short-Term Memory network, and the Convolutional-Recurrent Neural Network. This is done by constructing and labeling datasets, dividing them into training, validation, and test sets. Each network is then iteratively trained until the required number of iterations is reached or the model converges. The iteration process ends when the desired number of iterations is achieved, resulting in a trained network model and thus a pre-trained fault classification model.

[0080] Specifically, it could be:

[0081] The improved U-Net network dataset is based on visual images (resolution no less than 2048×1536 pixels), infrared thermal images, terahertz images, and ultrasonic images. For each image type, 500 original images are taken and expanded to 3000 images per class through data augmentation techniques such as translation (±10 pixels), scaling (0.8-1.2 times), rotation (0-360°), random occlusion (occlusion area ≤5%), and grayscale conversion, for a total of 12000 images. Pixel-level annotation is used, employing professional annotation tools (such as Label Me) to annotate defect areas such as cracks, delamination, and wear, achieving an accuracy rate of ≥99%. The annotations are cross-checked and confirmed by two aviation maintenance engineers.

[0082] Long Short-Term Memory (LSTM) network dataset: 3000 sets each of tire pressure and temperature time-series data were collected, with tire pressure data accuracy up to 100 Pa and temperature data accuracy up to 0.1℃. Each data set contains 1000 time steps (sampling frequency 1 Hz, duration 16 minutes and 40 seconds). Labeling is categorized as "normal / abnormal," with abnormal data labeled with fault type (e.g., abnormal temperature corresponding to seal wear, pressure drop corresponding to tire pressure leakage). Labeling follows aviation industry fault judgment standards (e.g., SAEARP4754A).

[0083] CNN-RNN hybrid model dataset: The above image data and time series data are combined to construct 3000 sets of multimodal samples. Each set of samples contains 1 image + 1 time series data, labeled with the fault type, location and severity (e.g., "fatigue crack - support root - length 1.8mm").

[0084] During the iterative optimization of the model, a "5-fold cross-validation" method was used, dividing the dataset into a training set (80%), a validation set (10%), and a test set (10%). The iterative optimization was divided into three stages:

[0085] Initial training phase: Train the training set with the Adam optimizer (learning rate 0.0004) for 50 epochs; monitor the validation set loss (cross-entropy loss), and if the loss does not decrease for 5 consecutive epochs, reduce the learning rate to 0.00005;

[0086] Fine-tune phase: Freeze the backbone network weights and train only the newly added attention layer (improved U-Net) and the temporal feature fusion layer (CNN-RNN fusion) for 20 epochs with a learning rate of 0.8 times the initial learning rate.

[0087] Generalization optimization phase: Introduce 10% of real aviation maintenance fault data (including unlabeled interference data) for adversarial training to improve the model's anti-interference ability. The iteration stops when the final test set accuracy is ≥98% and the recall rate is ≥97%.

[0088] The pre-trained fault classification model is executed during the inference phase as follows: based on at least one type of data in the state data and reference values ​​of parameters in the digital twin model, a suspected fault type is determined, which includes at least fatigue cracks, wear, and displacement anomalies; based on at least one of the remaining types of data in the state data and simulation results based on the digital twin model, the suspected fault is verified to determine the output fault type.

[0089] Furthermore, the fault classification model performs the following during the inference phase: 1) For at least one type of data in the state data, it uses one of its network models to identify the fault type, or, in combination with the reference values ​​of the corresponding parameters in the digital twin model, it determines the suspected fault type; 2) For at least one type of data in the remaining state data, it uses one of its network models to identify the fault type (at this time, the identified fault type can be a supplement to the relevant information of the fault type identified in 1), or it can be the same fault type identified in 1), or, in combination with the simulation results of the digital twin model, it verifies the suspected fault identified by the network model in 1) or 2) to determine the output fault type.

[0090] In addition, the combination identification or verification of digital twin models in 1) or 2) above is not a necessary process. That is, in 1), a suspected fault can be identified by using only one of its network models, or in 2), a suspected fault can be verified by using only one of its network models based on state data different from that in 1).

[0091] Similarly, the network models in 1) or 2) above do not have to exist simultaneously. That is, in 1), suspected faults can be identified only through the digital twin model, or in 2), suspected faults can be verified only through the digital twin model.

[0092] In summary, the fault classification model in this embodiment determines the fault type in two steps: using at least one type of data from the state data, and through reference values ​​of parameters in its network model and / or digital twin model, it identifies suspected faults; using at least one type of data from other types of data from the state data, and through its network model and / or digital twin-based simulation, it verifies the suspected faults to determine the output fault type.

[0093] like Figure 3 As shown, the acquired multi-dimensional state data is first input into the pre-trained fault classification model, which then performs the following steps in sequence: first layer, data classification and routing; second layer, data preprocessing; third layer, feature extraction and representation; fourth layer, algorithm model analysis; fifth layer, multi-modal fusion diagnosis; and sixth layer, fault diagnosis output.

[0094] In the first layer of data classification and routing operations, the status data is classified as follows: image data acquired by high-resolution cameras and built-in cameras in the warehouse; sensor time-series data acquired by displacement sensors, pressure sensors, vibration sensors and infrared thermal imagers; three-dimensional geometric data acquired by ultrasonic phased arrays and other instruments; and non-destructive testing data acquired by ultrasonic phased array detectors and terahertz imagers.

[0095] The second-layer data preprocessing operation includes denoising / enhancing / segmenting image data, filtering / normalizing sensor time-series data, point cloud registration of 3D geometric data, and signal enhancement / feature extraction of non-destructive testing data.

[0096] In the third layer of feature extraction and representation operations, visual feature vectors are extracted for image data, temporal feature vectors are extracted for sensor time-series data, geometric feature vectors are extracted for 3D geometric data, and defect feature vectors are extracted for non-destructive testing data.

[0097] In the fourth-layer algorithm model analysis, for image data, a convolutional neural network (CNN) is used for image defect recognition to identify defects such as surface cracks, corrosion, and wear, and to accurately measure their dimensions. For sensor time-series data, a recurrent neural network (RNN) or a long short-term memory network (LSTM) is used to analyze data trends and predict performance degradation and potential failures. For three-dimensional geometric data, a point cloud comparison algorithm is used to calculate geometric deviations. Through point cloud registration and other techniques, the actual scanned three-dimensional point cloud is compared with the digital twin model to calculate geometric deviations and identify deformation and displacement anomalies. For non-destructive testing data, an improved U-Net deep learning algorithm is used to segment ultrasonic / X-ray images, enhance defect signals, and achieve automatic identification and quantitative analysis of minute defects.

[0098] The calculation of single-point geometric deviation can be expressed as: In the formula, d k The geometric deviation between the k-th point in the actual landing gear point cloud and the corresponding point in the digital twin reference model is (xk ,y k ,z k (x) represents the actual scan coordinates. k ,y k ,z k () represents the coordinates of the baseline model, all in mm.

[0099] The statistics of the overall geometric deviation can be expressed as:

[0100] ;

[0101] In the formula, d represents the average geometric deviation of the target area (e.g., the base of a support column). max The maximum geometric deviation is M, where M is the total number of point clouds in the target region, and d is the maximum geometric deviation. k The values ​​represent single-point deviations (calculated using the single-point deviation formula above), and the units are all in mm.

[0102] In the fifth-layer multimodal fusion diagnosis process, the analysis results of the fourth-step algorithm are combined with at least one of the following methods: digital twin model comparison, multi-source data correlation analysis, abnormal logic cross-validation, and fault feature fusion to perform fault fusion diagnosis.

[0103] During the sixth-level fault diagnosis output process, the fault type is identified as crack / wear / deformation, etc., and the fault location is accurately located. The severity of the fault is quantified, such as crack length, depth, wear amount, deformation amount, etc.

[0104] Step 13: Based on the fault information output by the fault classification model and the preset fault analysis database, assess the risk level of the fault.

[0105] The Failure Mode and Effects Analysis (FMEA) database is used to assess the risk level of failures to landing gear function and flight safety, providing a basis for subsequent maintenance decisions.

[0106] The FMEA risk level RPN (Risk Priority Number) is determined by the product of the probability of a failure occurring, the severity of its consequences, and the difficulty of its detection. Specifically, it can be expressed as:

[0107] RPN = O × S × D, where O is the probability of failure (1-10 points, 1 = very low, 10 = very high), S is the severity of the consequences of failure (1-10 points, 1 = minor, 10 = fatal), and D is the difficulty of failure detection (1-10 points, 1 = easy to detect, 10 = difficult to detect). The RPN range is 1-1000.

[0108] The correspondence between RPN values ​​and risk levels is as follows: 1≤RPN≤100 indicates low risk; 101≤RPN≤300 indicates medium risk; 301≤RPN≤600 indicates high risk; and RPN>600 indicates extremely high risk.

[0109] Step 14: Based on the fault information, the assessed risk level, and the preset expert knowledge base, determine the fault repair plan.

[0110] The expert knowledge base stores maintenance standards, as well as landing gear maintenance manuals, historical maintenance cases, and best practices, among other maintenance reference information. Based on the type, location, severity, and risk level of the fault, relevant information is retrieved from the knowledge base for reasoning and decision-making.

[0111] For low-risk faults such as minor wear or early micro-cracks, generate targeted repair solutions, such as grinding, welding, and coating repair.

[0112] For high-risk failures such as severe cracks, structural deformation, or failure of critical components, a replacement plan is generated, specifying the model, quantity, and specifications of the parts that need to be replaced.

[0113] For systemic problems such as hydraulic system leaks and electrical connection failures, generate system troubleshooting and repair plans;

[0114] Based on the fault diagnosis results and risk assessment level, relevant maintenance strategies and cases are retrieved from the expert knowledge base, and reasoning and optimization are performed to generate a specific maintenance plan. The plan clearly defines the maintenance objectives, detailed operating procedures, required parts models and quantities, tool and equipment list, safety protection measures, quality inspection standards, and estimated working hours.

[0115] like Figure 4 As shown, the implementation of step 14 may specifically include: inputting a diagnostic result containing fault information and risk level, and performing the following steps in sequence based on the diagnostic result: first layer, fault information parsing; second layer, expert knowledge base retrieval; third layer, maintenance strategy formulation; fourth layer, maintenance plan generation; fifth layer, maintenance effect simulation; and sixth layer, plan optimization decision.

[0116] During the first-level fault information analysis process, the analysis and diagnosis results yield information such as fault type (crack / wear / deformation), precise coordinate location of the fault, quantitative indicators of severity, and specific risk level (low / medium / high / extremely high).

[0117] During the second-level expert knowledge base retrieval process, based on the analysis results, the system retrieves maintenance strategies that match the analysis results from industry standard maintenance manuals, historical maintenance case matching, and best practice experience bases, and also views the inventory information of the parts BOM (Bill of Materials) used in the maintenance strategies.

[0118] The third-level maintenance strategy development process includes repair solutions such as grinding / welding / coating, replacement solutions such as determining the model / quantity of replacement parts, system maintenance such as troubleshooting / debugging, and combined maintenance such as setting up combination strategies.

[0119] The process of generating a fourth-level maintenance plan includes detailed steps and operating procedures for generating the maintenance plan; a list of required tools and equipment; safety precautions and protection requirements during the maintenance process; and an estimated maintenance time and cost.

[0120] In the fifth-level modification effect simulation process, the effectiveness of the maintenance plan is verified by simulating the expected effects after maintenance based on the proposed maintenance plan. This includes using a digital twin model for pre-maintenance simulation, performance recovery prediction, life cycle assessment, and risk elimination effect verification. In the sixth-level plan optimization decision-making process, multiple maintenance plans are compared and analyzed, cost-benefit assessment is performed, resource availability is checked, and the optimal plan is selected and confirmed, resulting in the final determined maintenance plan.

[0121] The expected outcome is based on the performance recovery rate characterization of the digital twin model, and the repair is deemed successful when the performance recovery rate exceeds a set threshold. The set threshold can be 90%.

[0122] Specifically, the performance recovery rate after repair is calculated as follows:

[0123]

[0124] η represents the performance recovery rate (%). For digital twin simulation of post-repair performance parameters (such as vibration frequency after replacing the support column). If the standard performance parameters of the benchmark model (such as normal vibration frequency 12-15Hz) are η≥90%, the repair is deemed qualified.

[0125] In one feasible implementation, this embodiment also includes organizing and outputting the results of the entire testing and analysis process to generate a standardized testing report. The testing report integrates testing data, diagnostic results, risk assessments, and maintenance plans, outputting them in document and visual chart formats. The specific content of the testing report includes information on the tested object, testing time and environment, testing equipment and methods used, a summary of collected data, a list and detailed description of faults diagnosed by the fault classification model, risk assessment results, recommended maintenance plans, and related images and data charts. This report not only provides maintenance personnel with clear operational guidelines but also provides decision-making support for airline maintenance management departments, and serves as an important archive for the landing gear's entire lifecycle management. Furthermore, the data in the testing report can be used for digital twin models and optimizing fault detection algorithms, providing a more accurate benchmark and more intelligent analytical capabilities for subsequent testing.

[0126] The method of this embodiment will be described in detail below with reference to the specific implementation method.

[0127] Specific implementation method 1: Detect fatigue cracks inside the landing gear struts.

[0128] like Figure 5 As shown, for fatigue cracks inside the landing gear strut, the data detected by the ultrasonic phased array detector is used as the criterion, and the reference strut structure is determined based on the digital twin model, such as the material: 300M ultra-high strength steel with a wall thickness of 15mm; the high-incidence areas of fatigue cracks are located at the root of the strut and the flange connection.

[0129] During the inspection, an ultrasonic phased array detector (probe frequency 5MHz, array element number 64) emits ultrasound along a preset path to perform a fan-shaped scan of the landing gear strut. The scan covers the entire strut surface, focusing on high-incidence areas marked in the model, and acquires a large amount of ultrasonic echo signals and image data containing internal structural information. The acquired ultrasonic images are then denoised (using a wavelet transform denoising algorithm) and gain compensated to enhance the defect signal. Finally, the processed ultrasonic images are registered with corresponding positions in the digital twin model to establish a mapping relationship between image pixels and actual physical locations.

[0130] State data, including ultrasonic image data acquired by an ultrasonic phased array detector, is input into the fault classification model, enabling the fault classification model to execute:

[0131] For the ultrasonic image data in the non-destructive testing data, defect segmentation is performed using an improved U-Net to identify crack features;

[0132] The crack features are correlated and compared with the structural parameters at the corresponding crack location in the digital twin model to obtain the crack location and crack severity, thereby identifying suspected fatigue crack faults.

[0133] For the vibration data corresponding to the crack location in the physical parameter data, vibration anomalies are identified by long short-term memory network, and the fatigue crack fault of the landing gear is determined by combining the dynamic simulation results based on the digital twin model.

[0134] Specifically: The fault classification model loads the registered ultrasound image data and digital twin model, and performs core analysis: Defect segmentation of the ultrasound images is performed using an improved U-Net convolutional neural network—the network first extracts features from the echo signals in the image, identifying "abnormal regions" that differ from the echoes of normal metal tissue, and then determines the boundaries of suspected cracks through pixel-level segmentation, achieving a segmentation accuracy of 98.5%. Figure 6 As shown in the figure, the part highlighted in red is a suspected crack with a length of 2 mm and a width of 0.01 micrometers. The segmented crack features are compared with the strut structure parameters in the digital twin model. The digital twin model has pre-stored the design parameters of this type of landing gear strut (made of 300M ultra-high strength steel), including a wall thickness of 15 mm, a fatigue crack critical length of 0.02 mm, and a maximum allowable depth of 3 mm.

[0135] The fault classification model measures the geometric parameters of suspected cracks. In this inspection, an internal crack with a length of 1.8 mm and a depth of 2.5 mm was identified at the root of the support column (the "high-risk fatigue area" marked by the digital twin model). Its location completely coincides with the "force transmission path node" in the model, and its size exceeds the design allowable critical value. It is judged as a "structural anomaly", that is, a suspected fatigue crack fault is identified.

[0136] Specifically, the quantification of crack length L and depth H is as follows:

[0137]

[0138] In the formula, L is the crack length, H is the crack depth, and (x1, y1, z1) and (x2, y2, z2) are the crack lengths and depths. 2） The coordinates of the two ends of the crack after ultrasonic image segmentation are (mapped by pixel-physical size), and the unit is mm.

[0139] For a confirmed suspected fatigue crack fault, based on the vibration data of the support column collected by the vibration sensor and the dynamic simulation results of the digital twin model (the vibration frequency of the support column under normal conditions is 12-15Hz), it was found that the vibration frequency in this area abnormally dropped to 9.2Hz, which further verified the abnormal logic that "cracks lead to a decrease in structural stiffness". After eliminating misjudgment caused by signal interference, the fault classification model outputs landing gear fatigue crack fault.

[0140] The fault type is fatigue crack fault, which belongs to the landing gear fatal fault category (the risk level RPN base value in FMEA is 85), and the risk level is extremely high, requiring immediate repair.

[0141] The risk level was determined by a combination of factors, including “probability of failure (medium, due to the crack exceeding the critical length and accelerating propagation),” “severity of consequences (extremely high, involving flight safety),” and “difficulty of detection (low, already clearly located by ultrasound),” and was ultimately classified as “extremely high risk.”

[0142] In addition, regarding the impact analysis of fatigue crack failure: the root of the strut is the main load-bearing point during takeoff and landing. If the crack continues to expand, it will reach the fracture critical value within 30 takeoff and landing cycles (based on fatigue life simulation results of digital twin model), which may lead to landing gear failure and cause aircraft landing accidents.

[0143] Based on the fault information in this implementation method, the risk level is assessed, and expert knowledge base matching is initiated: Parts selection: According to the BOM list associated with the digital twin model, the original part model (LG-1234-01) of this type of support is automatically retrieved, and the inventory status is checked (the system is connected to the airline's spare parts management database, which shows that there are 2 parts in stock).

[0144] Maintenance steps: First, depressurize the landing gear hydraulic system and disconnect the hydraulic lines at the top and bottom of the strut (refer to the "Hydraulic Interface Assembly Drawing" in the digital twin model, specifying the line torque value as 35 N·m); use specialized lifting equipment (model HL-500) to remove the strut from the wing landing gear bay, avoiding collisions with the electrical wiring harness inside the bay (the model marks the wiring harness location coordinates, indicating "avoid the X=1200-1300mm, Y=800-900mm area during lifting"); install the new strut, adjusting the installation position according to the "Geometric Tolerance Requirements" (coaxiality ≤ 0.1mm) in the digital twin model, and tighten the connecting bolts to the specified torque using a torque wrench; reconnect the hydraulic lines and perform a sealing test (pressure 15MPa, pressure holding for 30min, leakage ≤ 0.5mL / min, refer to the "Hydraulic System Performance Parameters" in the model).

[0145] Tools and safety requirements: It is clearly stated that an ultrasonic thickness gauge (to verify the wall thickness of the new strut) and a hydraulic pump station (to test the sealing performance) must be used, and the landing gear electrical control circuit must be disconnected during operation to prevent malfunction.

[0146] Generate a "Landing Gear Strut Inspection and Repair Report". The core contents of the report include:

[0147] Detection data: Ultrasonic images (with crack location and size marked), vibration frequency comparison curves (normal vs. abnormal state), and screenshots of digital twin model comparison;

[0148] Diagnosis: Extremely high risk fatigue crack (location: root of the support column, parameters: 1.8mm × 2.5mm).

[0149] Repair plan: Complete replacement steps, parts models, tool list, and quality acceptance standards;

[0150] Risk warning: If not replaced in time, there is a risk of breakage after 30 take-off and landing cycles.

[0151] At the same time, the crack data, vibration signals, and maintenance records from this inspection are stored in the database to update the digital twin model. The model learns the crack propagation pattern and optimizes the detection weight of subsequent "high-risk fatigue areas," thereby increasing the identification speed of similar defects by 20%.

[0152] Specific implementation method 2: Detect abnormal temperature of landing gear retraction mechanism.

[0153] In this implementation, for the aircraft's nose landing gear retraction mechanism (including hydraulic actuators and locking mechanisms), an infrared thermal imager is used to capture abnormal temperatures and locate hydraulic system leaks.

[0154] The landing gear retraction mechanism in the digital twin model is based on the CAD assembly drawing of the nose landing gear retraction mechanism. A three-dimensional digital twin model is constructed, including a hydraulic actuator (model HY-456), a locking hook, and a connecting rod. Thermodynamic parameters are also integrated: the outer wall temperature of the actuator should be ≤55℃ during normal operation, the friction surface temperature of the locking mechanism should be ≤60℃, and the hydraulic oil operating temperature should be 40-50℃. The digital twin model specifically marks key components prone to heat generation, such as the actuator seal and the locking mechanism pin.

[0155] For abnormal temperatures in the landing gear retraction mechanism, data collected by an infrared thermal imager is used as the initial criterion. The infrared thermal imager (640×512 resolution, temperature range -50-150℃, accuracy ±2%) continuously captures images of the retraction mechanism (10fps). The acquired infrared thermal images are shown below. Figure 7 As shown, the temperature field changes during the extension and retraction process are recorded simultaneously; at the same time, the displacement sensor records the extension and retraction speed of the actuator cylinder (normally ≥100mm / s), and the pressure sensor monitors the hydraulic system pressure (normal extension and retraction pressure 18-20MPa).

[0156] The acquired infrared thermal images undergo data preprocessing and registration: "pseudo-color enhancement" and "temperature calibration" are performed on the infrared thermal images to eliminate the interference of ambient light (such as hangar lights) on temperature measurement; the geometric center of the actuator and locking mechanism is located through image recognition technology, and the corresponding parts of the infrared thermal images and digital twin models are aligned in coordinates to ensure that the temperature data corresponds one-to-one with the physical structure.

[0157] Input the state data, including the infrared thermal image, into the fault classification model, and then execute the fault classification model:

[0158] The landing gear status data, along with a digital twin model matching the current physical state of the landing gear, are input into a pre-trained fault classification model, causing the fault classification model to perform:

[0159] For the infrared thermogram data in the thermal detection data, temperature anomalies are identified by the improved U-Net, and the temperature anomaly areas are determined by combining the standard values ​​of the corresponding area temperature marked in the digital twin model; the infrared thermogram data characterizes the temperature changes during the landing gear retraction and extension process.

[0160] Within the same time period, for the displacement data representing the extension and retraction of the actuator cylinder in the physical parameter data of the temperature anomaly area, a long short-term memory network is used to identify and determine suspected wear faults of the actuator cylinder piston rod seal.

[0161] For the pressure data of the hydraulic system in the physical parameter data, pressure anomalies are identified by long short-term memory network, and combined with the hydraulic system flow field simulation results based on digital twin model, the wear failure of the piston rod seal of the actuator is determined.

[0162] Specifically: The fault classification model initiates the "time-series temperature analysis + multi-parameter correlation" logic, including temperature comparison and multi-parameter verification processes.

[0163] Temperature comparison: extraction such as Figure 8 The infrared thermal image corresponding to the temperature curve of the actuating cylinder sealing area shows that the temperature in this area rapidly increased from the initial 42℃ to 78℃ during the extension and retraction cycle, far exceeding the standard value of "≤55℃" marked by the digital twin model. Moreover, the temperature rise period completely coincides with the "actuating cylinder extension and retraction jamming" period in the stretching rate curve (displacement sensor data shows that the speed drops to 30mm / s during the jamming period).

[0164] Multi-parameter verification: Hydraulic system pressure data shows that the pressure fluctuation during the retraction and extension jamming period reaches ±3MPa (normal fluctuation ≤ ±0.5MPa). Combined with the "hydraulic system flow field simulation" of the digital twin model, which shows that when the seal wears, hydraulic oil leakage will lead to pressure loss, and at the same time, increased friction will cause temperature rise, it is completely consistent with the measured data. Other interfering factors such as "abnormal hydraulic oil viscosity" are eliminated, and it is determined to be "hydraulic system seal failure type anomaly".

[0165] An impact analysis was conducted on the identified abnormal wear of the piston rod seal in the actuator cylinder: the wear of the seal leads to hydraulic oil leakage, which in the short term will cause a slowdown in the retraction and extension speed, affecting takeoff and landing efficiency; in the long term, the leakage will cause the hydraulic system oil level to drop, causing "cavitation" in the actuator cylinder, and may even cause the locking mechanism to fail to lock reliably, posing a risk of accidental retraction and extension of the landing gear in the air.

[0166] Risk level assessment: Based on the FMEA database, the RPN value of "seal wear" is 65. Considering the "probability of occurrence (high, the seal has been used for 2000 landing cycles, exceeding 80% of the design life)" and "severity of consequences (medium, it does not directly endanger structural safety at present, but affects functional reliability)," the risk level is determined to be "medium risk" and repairs need to be completed within 10 landing cycles.

[0167] Generate maintenance plan: The maintenance decision and handling module generates a special plan for "sealing replacement".

[0168] Parts selection: Based on the BOM of the digital twin model, the sealing part model is determined to be HS-789 (fluororubber material, oil resistance temperature -20-120℃).

[0169] The specific repair steps include:

[0170] Lock the front landing gear in the "lower" position, depressurize the hydraulic actuator, and remove the dust cover at the end of the actuator.

[0171] Use a special puller (model TL-200) to remove the worn seal and clean the residual hydraulic oil and impurities in the seal groove (refer to the "Seal Groove Dimension Diagram" in the model to ensure there are no scratches on the bottom of the groove);

[0172] Apply special hydraulic oil (model MIL-PRF-83282) to lubricate the new seals and install them with the lip facing the pressure side (the installation direction of the seals is marked on the model to avoid reverse installation).

[0173] Reinstall the dust cover, start the hydraulic system to conduct a take-up and take-down test, verify that the temperature of the sealing area is ≤55℃ using an infrared thermal imager, and verify that the take-up and take-down speed has been restored to 110mm / s using a displacement sensor.

[0174] Cost and time: The estimated repair time is 2 hours, and the cost of parts is 800 yuan. There is no need to replace the core components, and the cost performance is better than replacing the entire actuator.

[0175] A test report is generated, which includes a "temperature anomaly time series curve" and a "seal installation diagram" (derived from the digital twin model). The wear data of the seal is entered into the database. The digital twin model learns the "temperature-pressure-life" correlation of the fault and optimizes the temperature monitoring threshold of the "seal area" for the next test (from 55℃ to 52℃), so as to detect potential wear earlier.

[0176] Specific implementation method 3: Detect delamination defects in composite material landing gear doors.

[0177] In this implementation, for the composite material landing gear door of a certain type of regional aircraft (the material is T800 carbon fiber / epoxy resin), terahertz imaging technology is used to detect internal delamination defects. This defect is a typical latent fault of composite materials and cannot be detected by traditional visual inspection.

[0178] The landing gear door section in the digital twin model is based on the CAD design drawings of the door, constructing a 3D model with a three-layer structure of "panel-honeycomb core-backplate". Key parameters are pre-stored in the model: panel thickness 2mm, honeycomb core height 10mm, and backplate thickness 1.5mm. Areas prone to delamination, such as "door edges" and "hinge connection areas", are also marked (these areas are prone to air bubbles during composite material molding, leading to delamination later). Simultaneously, the digital twin model integrates the propagation parameters of terahertz waves in this composite material (refractive index 1.6, attenuation coefficient 0.2dB / mm) for subsequent defect depth calculations.

[0179] For delamination defects in composite landing gear doors, data acquired by a terahertz imager was used as the initial criterion. The terahertz imager employed a 0.3THz terahertz imaging system (scanning resolution 0.1mm, imaging speed 5cm² / s) to perform a full-coverage scan of the landing gear door. Terahertz waves possess the characteristics of "penetrating non-metallic materials and being sensitive to changes in dielectric constant." When encountering the delamination interface within the composite material (where the dielectric constants of air and resin differ significantly), a strong reflection signal is generated. The system simultaneously records the time delay and amplitude data of the terahertz reflected waves, generating a two-dimensional imaging matrix of "reflection signal-position."

[0180] The acquired terahertz imaging data underwent data preprocessing and registration: background noise removal (using an average filtering algorithm) and signal amplitude normalization were performed on the terahertz imaging data to eliminate the interference of oil stains on the reflected signal on the hatch surface; the terahertz imaging matrix was coordinately registered with the hatch structure of the digital twin model using the positioning hole on the edge of the hatch (φ5mm hole marked in the CAD drawing) as a reference point to ensure that each imaging pixel corresponds to the precise position in the model (e.g., "hinge connection area X=500mm, Y=300mm").

[0181] Input the state data, which includes terahertz imaging data, into the fault classification model, and then execute the fault classification model:

[0182] The landing gear status data, along with a digital twin model matching the current physical state of the landing gear, are input into a pre-trained fault classification model, causing the fault classification model to perform:

[0183] For the terahertz imaging data in the non-destructive testing data, a convolutional neural network is used to identify the abnormal material layering regions in the landing gear doors.

[0184] Based on the terahertz wave propagation parameters in the digital twin model, the delamination area is calculated to identify suspected delamination defects in composite materials.

[0185] For the door locking pressure data in the physical parameter data, a recurrent neural network is used to identify pressure anomalies in order to determine the delamination defects of the composite material.

[0186] Specifically, the fault classification model uses a fusion algorithm of "deep learning + electromagnetic wave simulation" to perform defect identification, deep calculation and verification processes.

[0187] Defect identification: The terahertz imaging data was classified using a ResNet-50 convolutional neural network. During network training, more than 1,000 sets of terahertz image samples of "normal composite materials" and "delamination defects" were introduced, which can automatically identify "delamination regions" with abnormal reflected signals. In this detection, two delamination defects were identified in the door hinge connection area (the "high stress area" marked on the model). Figure 9 (as shown)

[0188] Depth Calculation: Combining the terahertz wave propagation parameters in the digital twin model, the fault classification model calculates the layering depth through the "reflected wave time delay"—the first layer is located 1.8mm below the "panel-cell core" interface (close to the panel thickness of 2mm, belonging to the surface layer), and the second layer is located 2.2mm above the "cell core-backsheet" interface (belonging to the deep layer). The areas of the two layers are 15mm×8mm and 20mm×12mm, respectively, both exceeding the design allowable standard of "maximum layering area of ​​10mm×10mm".

[0189] Calculation of delamination depth in composite materials:

[0190]

[0191] Note: h is the layer depth (mm), c = 3 × 10 8 m / s is the propagation speed of terahertz waves in a vacuum, Δt is the time delay of the reflected wave (s), n=1.6 is the refractive index of T800 carbon fiber / epoxy resin for terahertz waves, and the denominator "2" corresponds to the round-trip path of the wave.

[0192] Verification: By analyzing the "locking pressure data" collected by the pressure sensor when the hatch is closed, it was found that the locking pressure in the delaminated area was 15% lower than that in the normal area (model simulation showed that delamination leads to a decrease in the local stiffness of the hatch, making it prone to deformation when locked). This further verified the impact of the defect on the hatch function. Ultimately, it was determined to be a delamination defect in the composite material.

[0193] Analysis of the impact of composite material delamination defects: As a protective component of the landing gear, the delamination of the door will lead to a decrease in local strength. During flight, when the door is subjected to airflow load, the delamination area may expand, causing the door to crack or even fall off, and at the same time affecting the "door synchronous opening and closing" function when the landing gear is retracted and extended.

[0194] Risk level assessment: Referring to the FMEA database, the RPN value for "composite material delamination" is 70. Combined with "probability of occurrence (medium, the door has been used for 1500 flight hours, close to 60% of the fatigue life of the composite material)" and "seriousness of consequences (high, door detachment may endanger ground personnel or aircraft)", the risk level is determined to be "high risk" and repairs must be completed within 5 flight days.

[0195] A repair plan for "vacuum-assisted resin injection repair" was generated.

[0196] Material selection: Based on the composite material maintenance specifications associated with the digital twin model, “T800 carbon fiber cloth (unidirectional, thickness 0.2mm)” and “epoxy resin adhesive (model EP-610, curing temperature 80℃)” matching the hatch substrate were selected.

[0197] The specific repair steps include:

[0198] Mark the layered areas (refer to the coordinate annotations of the digital twin model), and use sandpaper (400 grit) to polish the surface to remove the coating and oxide layer;

[0199] Drill φ1mm injection holes (5mm spacing) in the layered areas, inject low viscosity resin through the injection holes, and at the same time use a vacuum bag (vacuum degree -0.095MPa) to evacuate the vacuum to ensure that the resin fills the gaps between the layers.

[0200] Lay out carbon fiber cloth (cut according to the layer area, 10mm larger than the layer area), cover it with a vacuum bag again, and put it into a heating oven (according to the model's recommended "80℃×2h" curing process) for curing;

[0201] After curing, remove the vacuum bag, smooth the repaired area, and use a terahertz imaging instrument to check the repair effect to ensure that the layers are completely filled.

[0202] Quality acceptance: After repair, a "tensile test" (refer to the "composite material tensile strength standard 1800MPa" in the reference model) is required to ensure that the strength of the repaired area reaches more than 90% of the original structure.

[0203] A test report is generated, which includes a "terahertz imaging defect map" and a "repair process parameter table" (derived from the maintenance specifications of the digital twin model). The terahertz data and strength test data before and after repair are stored in the database. By learning the "formation-expansion-repair" process of this layered defect, the digital twin model optimizes the subsequent inspection frequency of "composite component" (from "once every 300 flight hours" to "once every 250 flight hours"), thereby improving the ability to detect early defects.

[0204] This application embodiment acquires multi-dimensional real-time status data of the landing gear using various high-precision sensors and imaging devices, and inputs this data along with a digital twin model into a fault classification model to obtain accurate landing gear fault information. Since the digital twin model is built based on the original design and manufacturing data of the landing gear and is continuously updated based on historical and operational data, the fault classification model, based on the analysis results of real-time data and the digital twin model matching the current physical state of the landing gear, can accurately identify faults. Furthermore, this embodiment automatically identifies potential faults such as wear, cracks, and abnormal displacement, assesses the risk level, and generates detailed inspection reports and maintenance plans. This system achieves closed-loop management of the entire process from fault prevention and accurate inspection to efficient maintenance, significantly improving the safety, accuracy, and efficiency of aircraft landing gear maintenance and effectively reducing aviation operational risks. In addition, this application, through multi-dimensional data acquisition, a real-time updated digital twin model, and cross-validation logic, controls the fault location error within a small range, and the fault identification types cover more than 95% of typical landing gear faults, significantly outperforming existing technologies.

[0205] Example 2:

[0206] To address the problem of low accuracy in detecting aircraft landing gear faults, and based on the same inventive concept as Embodiment 1, this application also provides an aircraft landing gear fault detection system.

[0207] The specific structural diagram of the system is as follows: Figure 10 As shown, it includes the following functional units 1001-1004:

[0208] The data acquisition unit 1001 is used to acquire landing gear status data for landing gear fault diagnosis; the status data includes at least visual image data of the landing gear, non-destructive testing data, thermal testing data and various physical parameter data.

[0209] The fault determination unit 1002 is used to input the landing gear status data and a digital twin model matching the current physical state of the landing gear into a pre-trained fault classification model to obtain the landing gear fault information output by the fault classification model; the digital twin model is a model that matches the physical entity of the landing gear, constructed based on the original design and manufacturing data of the landing gear and continuously updated based on historical and actual operation data; the fault information includes the fault type determined based on the status data, and the fault location and fault severity determined based on the status data and the digital twin model.

[0210] The fault determination unit is specifically used to: determine the suspected fault type based on at least one type of data in the state data and the reference value of the parameter in the digital twin model, wherein the fault type includes at least fatigue crack, wear and displacement anomaly; and verify the suspected fault based on at least one of the other types of data in the state data and the simulation results based on the digital twin model to determine the output fault type.

[0211] In one specific implementation, when identifying fatigue crack faults, the fault determination unit is specifically used for:

[0212] For the ultrasonic image data in the non-destructive testing data, defect segmentation is performed using an improved U-Net to identify crack features;

[0213] The crack features are correlated and compared with the structural parameters at the corresponding crack location in the digital twin model to obtain the crack location and crack severity, thereby identifying suspected fatigue crack faults.

[0214] For the vibration data corresponding to the crack location in the physical parameter data, vibration anomalies are identified by long short-term memory network, and the fatigue crack fault of the landing gear is determined by combining the dynamic simulation results based on the digital twin model.

[0215] In one specific implementation, when identifying wear faults in the piston rod seal of the actuator, the fault determination unit is specifically used for:

[0216] For the infrared thermogram data in the thermal detection data, temperature anomalies are identified by the improved U-Net, and the temperature anomaly areas are determined by combining the standard values ​​of the corresponding area temperature marked in the digital twin model; the infrared thermogram data characterizes the temperature changes during the landing gear retraction and extension process.

[0217] Within the same time period, for the displacement data representing the extension and retraction of the actuator cylinder in the physical parameter data of the temperature anomaly area, a long short-term memory network is used to identify suspected wear faults of the actuator cylinder piston rod seal.

[0218] For the pressure data of the hydraulic system in the physical parameter data, pressure anomalies are identified by long short-term memory network, and the wear failure of the piston rod seal of the actuator is determined by combining the hydraulic system flow field simulation results based on the digital twin model.

[0219] In one specific implementation, when identifying delamination defects in composite materials, the fault determination unit is specifically used for:

[0220] For the terahertz imaging data in the non-destructive testing data, a convolutional neural network is used to identify the abnormal material layering regions in the landing gear doors.

[0221] Based on the terahertz wave propagation parameters in the digital twin model, the delamination area is calculated to identify suspected delamination defects in composite materials.

[0222] For the door locking pressure data in the physical parameter data, a recurrent neural network is used to identify pressure anomalies and determine composite material delamination defects.

[0223] The fault risk assessment unit 1003 assesses the risk level of a fault based on the fault information output by the fault classification model and a preset fault analysis database.

[0224] The fault risk assessment unit is specifically used to determine the risk level based on the product of the probability of a fault occurring, the severity of the fault's consequences, and the difficulty of fault detection.

[0225] The maintenance decision unit 1004 determines the maintenance plan for the fault based on the fault information, the assessed risk level, and a preset expert knowledge base; the expert knowledge base stores maintenance standards and maintenance reference information.

[0226] The detection system in this embodiment further includes: simulating the expected effect after performing maintenance based on the maintenance plan to verify the effectiveness of the maintenance plan. The expected effect is characterized by the performance recovery rate of the digital twin model, and the maintenance is deemed qualified when the performance recovery rate exceeds a set threshold.

[0227] This embodiment acquires multi-dimensional real-time status data of the landing gear using various high-precision sensors and imaging devices, and inputs this data along with a digital twin model into a fault classification model to obtain accurate landing gear fault information. Since the digital twin model is built upon the original design and manufacturing data of the landing gear and is continuously updated based on historical and operational data, the fault classification model, based on the analysis results of real-time data and the digital twin model matching the current physical state of the landing gear, can accurately identify faults. Furthermore, this embodiment automatically identifies potential faults such as wear, cracks, and abnormal displacement, assesses the risk level, and generates detailed inspection reports and maintenance plans. This system achieves closed-loop management of the entire process from fault prevention and accurate inspection to efficient maintenance, significantly improving the safety, accuracy, and efficiency of aircraft landing gear maintenance and effectively reducing aviation operational risks.

[0228] Based on the same inventive concept as the foregoing embodiments of this application, this application also provides a computing device.

[0229] like Figure 11 As shown, the computing device includes a memory 1101 and a processor 1102. The memory 1101 can be configured to store various other data to support operation on the electronic device. Examples of such data include instructions for any application or method used to operate on the electronic device. The memory 1101 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0230] The processor 1102, coupled to the memory 1101, is used to execute the computer program stored in the memory 1101 to perform the aircraft landing gear failure detection method described in the foregoing embodiments.

[0231] When processor 1102 executes the computer program to perform a method for detecting aircraft landing gear failures, it acquires multi-dimensional real-time status data of the landing gear through various high-precision sensors and imaging devices, and inputs this data along with a digital twin model into a fault classification model to obtain accurate landing gear failure information. Since the digital twin model is built based on the original design and manufacturing data of the landing gear and is continuously updated based on historical and operational data, the fault classification model, based on the analysis results of real-time data and the digital twin model matching the current physical state of the landing gear, can accurately identify failures. Furthermore, this embodiment automatically identifies potential failures such as wear, cracks, and abnormal displacement, assesses the risk level, and generates detailed inspection reports and maintenance plans. This system achieves closed-loop management of the entire process from failure prevention and accurate detection to efficient maintenance, significantly improving the safety, accuracy, and efficiency of aircraft landing gear maintenance and effectively reducing aviation operational risks.

[0232] When the processor 1102 executes the computer program in the memory 1101, in addition to the functions described above, it can also perform other functions, as detailed in the descriptions of the preceding embodiments.

[0233] Furthermore, such as Figure 11 As shown, the computing device also includes other components such as a display 1104, a communication component 1103, a power supply component 1105, and an audio component 1106. Figure 11 The diagram only shows some components and does not mean that the computing device includes only these components. Figure 11 The components shown.

[0234] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a computer, can implement the methods provided in the above embodiments.

[0235] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0236] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0237] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of detecting a failure of an aircraft landing gear, characterized in that, The method comprises: acquiring landing gear state data for landing gear fault judgment; the state data at least includes visual image data, non-destructive testing data, thermal detection data and various physical parameter data of the landing gear; inputting the state data of the landing gear, reference values of corresponding parameters in a digital twin model matching the current entity state of the landing gear and simulation result data of the digital twin model into a pre-trained fault classification model to obtain fault information of the landing gear output by the fault classification model; the fault classification model is a fusion model comprising an improved U-Net for image segmentation, a long short-term memory network for time series data analysis and a convolution-recurrent neural network for multi-modal fusion; the digital twin model is a model matching the physical entity of the landing gear, which is constructed based on original design and manufacturing data of the landing gear and continuously updated based on historical and actual operation data; the fault information includes a fault type determined based on the state data and a fault location and a fault severity determined based on the state data and the digital twin model; based on the fault information output by the fault classification model and a pre-set fault analysis database, evaluating a risk level of the fault; based on the fault information, the evaluated risk level and a pre-set expert knowledge base, determining a maintenance scheme for the fault; the expert knowledge base stores maintenance standards and maintenance reference information; the method further comprises: inputting the state data of the landing gear and the digital twin model matching the current entity state of the landing gear into the pre-trained fault classification model, so that the fault classification model performs: based on at least one type of data in the state data and reference values of parameters in the digital twin model, determining a suspected fault type, the fault type at least including fatigue cracks, wear and displacement abnormalities; based on at least one type of data in the remaining types of data and based on simulation results of the digital twin model, verifying the suspected fault to determine the output fault type; for the analysis result of the fault classification model, at least one of digital twin model comparison, multi-source data correlation analysis, abnormal logic cross verification and fault feature fusion mode is used for fault fusion diagnosis to determine the fault location and the fault severity.

2. The method of detecting an aircraft landing gear failure according to claim 1, wherein, if the fault type includes fatigue cracks, inputting the state data of the landing gear, reference values of corresponding parameters in the digital twin model matching the current entity state of the landing gear and simulation result data of the digital twin model into the pre-trained fault classification model, so that the fault classification model performs: for ultrasonic image data in the non-destructive testing data, defect segmentation is performed by the improved U-Net to identify crack features; associating and comparing the crack features with structure parameters at corresponding crack positions of the digital twin model to obtain crack positions and crack severity to determine suspected fatigue crack faults; for vibration data corresponding to the crack positions in the physical parameter data, vibration abnormalities are identified by the long short-term memory network, and the fatigue crack fault of the landing gear is determined in combination with the dynamic simulation result based on the digital twin model. The fatigue crack failure of the landing gear is determined based on the dynamic simulation result of the digital twin model, including determining crack area vibration frequency anomaly based on the dynamic simulation result of the digital twin model; Verify the abnormal logic that the crack causes the stiffness to decrease, and exclude the false judgment caused by signal interference to determine the fatigue crack failure of the landing gear.

3. The method of detecting an aircraft landing gear failure according to claim 1, wherein, The failure type includes cylinder piston rod seal wear, The state data of the landing gear, and the reference value of the corresponding parameter in the digital twin model matching the current entity state of the landing gear and the simulation result data of the digital twin model are input into the pre-trained fault classification model, so that the fault classification model executes: For the infrared thermal image data in the thermal detection data, the temperature anomaly is identified by the improved U-Net, and the temperature anomaly area is determined by combining the standard value of the corresponding area temperature labeled in the digital twin model; the infrared thermal image data represents the temperature change of the landing gear during the landing gear retraction process; For the displacement data representing the cylinder extension and retraction in the physical parameter data of the temperature anomaly area in the same time period, a long short-term memory network is used to identify suspected cylinder piston rod seal wear failure; For the pressure data of the hydraulic system in the physical parameter data, a long short-term memory network is used to identify pressure anomalies, and the cylinder piston rod seal wear failure is determined by combining the hydraulic system flow field simulation result based on the digital twin model; The hydraulic system flow field simulation result based on the digital twin model is used to determine the cylinder piston rod seal wear failure, including that when the seal wears out and the friction increases, the temperature rises, and the hydraulic system seal failure anomaly is determined, and the cylinder piston rod seal wear failure is determined.

4. The method of detecting an aircraft landing gear failure according to claim 1, wherein, The failure type includes composite material delamination defect, then, The state data of the landing gear, and the reference value of the corresponding parameter in the digital twin model matching the current entity state of the landing gear and the simulation result data of the digital twin model are input into the pre-trained fault classification model, so that the fault classification model executes: For the terahertz imaging data in the non-destructive testing data, a convolutional neural network is used to identify the material anomaly delamination area existing in the landing gear door; Based on the terahertz wave propagation parameters in the digital twin model, the delamination area is calculated to determine the suspected composite material delamination defect; For the door locking pressure data in the physical parameter data, a recurrent neural network is used to identify pressure anomalies to determine the composite material delamination defect.

5. The method of detecting an aircraft landing gear failure according to claim 1, wherein, The risk level of the failure is determined based on the product of the possibility of the failure, the severity of the consequences of the failure, and the difficulty of detecting the failure.

6. The method of detecting an aircraft landing gear failure according to claim 1, wherein, The method further includes simulating the expected effect after performing maintenance based on the maintenance scheme to verify the effectiveness of the maintenance scheme; the expected effect is represented by the performance recovery rate of the digital twin model, and when the performance recovery rate exceeds a set threshold, it is determined that the maintenance is qualified.

7. An aircraft landing gear fault detection system characterised in that, It includes: A data acquisition unit for acquiring landing gear state data for landing gear failure judgment; The state data at least includes visual image data, non-destructive testing data, thermal detection data and various physical parameter data of the landing gear; The fault determination unit inputs the state data of the landing gear and reference values of corresponding parameters in the digital twin model matching the current entity state of the landing gear and simulation result data of the digital twin model into a pre-trained fault classification model to obtain fault information of the landing gear output by the fault classification model; The fault classification model is a fusion model including an improved U-Net for image segmentation, a long short-term memory network for time series data analysis and a convolution-recurrent neural network for multi-modal fusion; the digital twin model is a model matching the physical entity of the landing gear, which is constructed based on original design and manufacturing data of the landing gear and continuously updated based on historical and actual operation data; the fault information includes a fault type determined based on the state data and a fault location and a fault severity determined based on the state data and the digital twin model; The fault risk assessment unit assesses the risk level of the fault based on the fault information output by the fault classification model and a pre-set fault analysis database; The repair decision unit determines a repair scheme of the fault based on the fault information and the assessed risk level and a pre-set expert knowledge base; The expert knowledge base stores repair standards and repair reference information; The system further includes: inputting the state data of the landing gear and the digital twin model matching the current entity state of the landing gear into a pre-trained fault classification model, so that the fault classification model performs: Determining a suspected fault type based on at least one type of data in the state data and reference values of parameters in the digital twin model, the fault type at least including fatigue cracks, wear and displacement abnormalities; Verifying the suspected fault based on at least one type of data in the remaining types of data in the state data and simulation results based on the digital twin model to determine the output fault type; For the analysis result of the fault classification model, at least one of digital twin model comparison, multi-source data correlation analysis, abnormal logic cross verification and fault feature fusion mode is used for fault fusion diagnosis to determine the fault location and the fault severity.

8. An electronic device, comprising: Comprise: A processor, a memory and a program or instructions stored on the memory and executable on the processor, which implement the steps of the method of any one of claims 1-6 when executed by the processor.

9. A storage medium, characterized by Comprise: The storage medium stores a program or instructions, which implement the steps of the method of any one of claims 1-6 when executed by a processor.

Images