Space-time feature mining and risk assessment method, system and device of flight operation data and storage medium

Abstract

The application relates to the technical field of operation state monitoring and risk assessment, and provides a kind of flight operation data spatiotemporal feature mining and risk assessment method, system, equipment and storage medium, solve the problem of low early identification accuracy of mobile device composite fault.The method of the application comprises: obtaining flight operation data and optical signal data of a mobile device, the flight operation data comprising component state data and environmental parameter data; according to the optical signal data, combined with the structural deformation field of the mobile device, a strain thermal map of the structure surface is formed; the strain thermal map, component state data and environmental parameter data are spatiotemporally aligned and fused, and spatiotemporal features are extracted from the fusion result by enhancing the attention mechanism; a kernel fuzzy C-means clustering operation is performed on the spatiotemporal features to identify potential fault mode combinations, and a flight risk assessment result is determined based on the potential fault mode combinations.The application improves the early identification accuracy of mobile device composite fault.

Description

Technical Field

[0001] This application relates to the field of mobile device operation status monitoring and risk assessment technology, and in particular to a method, system, device and storage medium for spatiotemporal feature mining and risk assessment of flight operation data. Background Technology

[0002] Under complex operating conditions, the operational safety of mobile devices depends on real-time monitoring and risk assessment of multi-source data. It is necessary to accurately capture the coupling effects of structural deformation, component status and environmental factors in order to identify potential compound failure modes and assess the risk level.

[0003] Currently, a monitoring method based on multi-sensor data fusion uses a fixed-weight data overlay method, combined with traditional clustering algorithms to classify the operating status, and determines the risk level by setting a preset threshold.

[0004] This scheme relies on static weight allocation, which makes it difficult to adapt to dynamically changing operating conditions, resulting in insufficient sensitivity to complex faults. At the same time, traditional clustering methods have limited ability to process nonlinearly separable data, and the accuracy of risk assessment is limited by the rationality of the preset threshold. Summary of the Invention

[0005] This application provides a method, system, device, and storage medium for spatiotemporal feature mining and risk assessment of flight operation data, in order to solve the problem of low accuracy in early identification of complex faults of mobile devices in the prior art.

[0006] Firstly, this application provides a method for spatiotemporal feature mining and risk assessment of flight operation data, including:

[0007] Acquire flight operation data and optical signal data of the mobile device, wherein the flight operation data includes component status data and environmental parameter data;

[0008] Based on the optical signal data and the structural deformation field of the mobile device, a strain thermogram of the structural surface is formed.

[0009] The strain thermogram, the component state data, and the environmental parameter data are spatiotemporally aligned and fused, and spatiotemporal features are extracted from the fusion result through an enhanced attention mechanism.

[0010] Kernel fuzzy C-means clustering is performed on the spatiotemporal features to identify potential fault mode combinations, and flight risk assessment results are determined based on the potential fault mode combinations.

[0011] Optionally, performing kernel fuzzy C-means clustering on the spatiotemporal features to identify potential failure mode combinations, and determining the flight risk assessment result based on the potential failure mode combinations, includes:

[0012] The spatiotemporal features are mapped to the kernel space to generate a feature vector set. The kernel space is an implicit feature space generated by kernel function transformation in kernel fuzzy C-means clustering.

[0013] Initialize the cluster centers of the feature vector set, and form a set of cluster groups based on the cluster centers and the feature vector set;

[0014] Extract the core feature combinations of each cluster in the cluster set as potential fault mode combinations;

[0015] Based on a preset risk level comparison table, the potential failure mode combinations are mapped to flight risk assessment results.

[0016] Optionally, forming a cluster group set based on the cluster centroids and the feature vector set includes:

[0017] The positions of the cluster centers are iteratively optimized using a genetic algorithm.

[0018] Based on the optimized position, a correlation strength matrix is ​​generated using a preset membership function;

[0019] Based on the index of the maximum value of the row vector in the association strength matrix, the final clustering result of each feature vector in the feature vector set is determined;

[0020] Based on the final clustering results, all feature vectors are grouped according to their respective cluster centers to form cluster groups corresponding to each cluster center. All cluster groups are then integrated to form a cluster group set.

[0021] Optionally, the step of generating a correlation strength matrix based on the optimized position using a preset membership function includes:

[0022] Based on the optimized position, calculate the similarity metric from each feature vector in the feature vector set to each cluster center point in the kernel space;

[0023] For each feature vector, all similarity metrics are combined to form a similarity vector.

[0024] Input the similarity vector into the membership function to output the association strength value of the feature vector with each cluster center point;

[0025] Aggregate the association strength values ​​of all feature vectors to construct an association strength matrix between each feature vector and each cluster centroid.

[0026] Optionally, the step of spatiotemporally aligning and fusing the strain thermogram, the component state data, and the environmental parameter data, and extracting spatiotemporal features from the fusion result through an enhanced attention mechanism, includes:

[0027] The strain thermogram, the component status data, and the environmental parameter data are each independently sliced ​​according to a unified timestamp to obtain three types of time slice groups at the same time.

[0028] Based on the structural space model of the mobile device, the three types of time slice groups are uniformly mapped to a preset three-dimensional spatial grid coordinate system;

[0029] In the three-dimensional spatial grid coordinate system, three types of time slice groups within the same spatial grid cell are dynamically weighted and fused to generate a fused grid tensor;

[0030] Spatiotemporal attention weighting is performed on the fused mesh tensor to generate spatiotemporal features.

[0031] Optionally, performing a spatiotemporal attention weighting operation on the fused mesh tensor to generate spatiotemporal features includes:

[0032] The fused mesh tensor is divided into multiple time windows along the time dimension, and time attention weights are generated based on the changing trend of mesh data within each time window.

[0033] In the spatial dimension, spatial attention weights are generated based on the magnitude of data variation between grid cells in the fused grid tensor.

[0034] The fused mesh tensor is decomposed into multiple feature channels according to the data source type, and channel attention weights are generated based on the covariance between the feature channels.

[0035] The temporal attention weights, spatial attention weights, and channel attention weights are multiplied by tensors to construct a three-dimensional attention weight tensor.

[0036] The spatiotemporal features are output by multiplying the three-dimensional attention weight tensor element-wise with the fused mesh tensor.

[0037] Optionally, the step of forming a strain thermogram of the structural surface based on the optical signal data and the structural deformation field of the mobile device includes:

[0038] The phase offset distribution is analyzed from the optical signal data;

[0039] Based on the pre-set fiber optic sensor network topology in the structural deformation field of the mobile device, the phase offset is bound to the corresponding spatial location point;

[0040] Based on each spatial location point and its corresponding phase offset, the local deformation is calculated using a preset phase and deformation transformation model.

[0041] Analyze the differences in local deformation between adjacent spatial locations to determine the strain intensity distribution on the structural surface;

[0042] The strain intensity distribution is transformed into a data visualization according to a preset color gradient rule to generate a strain thermogram of the structural surface.

[0043] Secondly, this application provides a system for spatiotemporal feature mining and risk assessment of flight operation data, including:

[0044] The acquisition module is used to acquire flight operation data and optical signal data of the mobile device, wherein the flight operation data includes component status data and environmental parameter data;

[0045] A forming module is used to form a strain thermogram of the structural surface based on the optical signal data and the structural deformation field of the mobile device.

[0046] The extraction module is used to perform spatiotemporal alignment and fusion of the strain thermogram, the component state data and the environmental parameter data, and extract spatiotemporal features from the fusion result through an enhanced attention mechanism;

[0047] The identification module is used to perform kernel fuzzy C-means clustering on the spatiotemporal features to identify potential fault mode combinations, and to determine the flight risk assessment result based on the potential fault mode combinations.

[0048] Thirdly, this application provides a computing device, including a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a method for spatiotemporal feature mining and risk assessment of flight operation data as described in any of the first aspects.

[0049] Fourthly, this application provides a computer storage medium storing computer program instructions thereon, which, when executed by a processor, implement a method for spatiotemporal feature mining and risk assessment of flight operation data as described in any one of the first aspects.

[0050] This application provides a method for spatiotemporal feature mining and risk assessment of flight operation data. The method includes: acquiring flight operation data and optical signal data of a mobile device, wherein the flight operation data includes component status data and environmental parameter data; forming a strain heat map of the structural surface based on the optical signal data and the structural deformation field of the mobile device; performing spatiotemporal alignment and fusion of the strain heat map, the component status data, and the environmental parameter data, and extracting spatiotemporal features from the fusion result through an enhanced attention mechanism; performing kernel fuzzy C-means clustering on the spatiotemporal features to identify potential fault mode combinations; and determining the flight risk assessment result based on the potential fault mode combinations.

[0051] The technical solution provided in this application has the following beneficial effects:

[0052] This application achieves synchronous acquisition of multi-source heterogeneous data, providing a complete data foundation for comprehensive risk assessment. It converts optical signals into high-precision deformation distributions, visually presenting areas of localized stress concentration in the structure. Through dynamic weighted fusion and attention mechanisms, it highlights key characteristics of the coupling effect between the environment, structure, and equipment. It identifies potential combinations of features from complex faults, achieving precise matching of risk levels.

[0053] Furthermore, this application also maps spatiotemporal features to a high-dimensional kernel space to generate a feature vector set, optimizes cluster center points to form group sets, extracts the core feature combination of each group as the fault mode, and finally outputs the risk assessment result based on preset rules.

[0054] Furthermore, by enhancing the separation capability of nonlinear features through kernel space transformation and combining it with fuzzy clustering to accurately capture the boundary uncertainty of fault modes, reliable identification and risk quantification of compound faults under complex working conditions can be achieved.

[0055] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

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

[0057] Figure 1 A flowchart illustrating a method for spatiotemporal feature mining and risk assessment of flight operation data, provided as an embodiment of this application;

[0058] Figure 2 A schematic diagram of the structure of a system for spatiotemporal feature mining and risk assessment of flight operation data provided in this application embodiment;

[0059] Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0060] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0061] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0062] Existing methods for assessing operational risks of mobile devices, which employ fixed-weight fusion of multi-source data combined with traditional clustering analysis, have significant shortcomings. Firstly, fixed weights struggle to adapt to the dynamic coupling between structural deformation, component vibration, and environmental factors, resulting in insufficient sensitivity to sudden, complex failures. Secondly, traditional clustering has limited ability to distinguish nonlinear features, easily categorizing weakly correlated features, leading to significant discrepancies between risk assessment results and actual operating conditions. These problems stem from insufficient adaptability of data processing methods to complex operational scenarios.

[0063] To address the aforementioned limitations, this invention proposes a method for spatiotemporal feature mining and risk assessment of flight operation data. Its innovation lies in: real-time analysis of structural deformation distribution using optical signals, combined with spatiotemporal alignment and fusion of component states and environmental parameters, and extraction of key coupling features using adaptive weight allocation and attention mechanisms; furthermore, an improved fuzzy clustering algorithm is employed to accurately identify fault mode combinations in a high-dimensional feature space, ultimately matching preset rules to output risk levels. This method overcomes the limitations of fixed weights and linear analysis, and through the synergy of dynamic fusion and nonlinear clustering, it not only improves the ability to capture complex faults but also ensures the accuracy of risk assessment, effectively solving the problem of poor adaptability of existing technologies to complex operating conditions.

[0064] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0065] Figure 1 A flowchart illustrating a method for spatiotemporal feature mining and risk assessment of flight operation data provided in this application embodiment is shown below. Figure 1 As shown, the method includes:

[0066] Step 101: Acquire flight operation data and optical signal data of the mobile device, wherein the flight operation data includes component status data and environmental parameter data.

[0067] In step 101, flight operation data includes component status data (such as engine vibration amplitude and control surface deflection angle) and environmental parameter data (such as wind speed, temperature, and air pressure). Components specifically refer to key functional components of mobile equipment that are directly related to flight safety, including but not limited to core mechanical components that generate vibration / displacement signals, such as engine rotors, wing control surfaces, and landing gear hydraulic systems. Wing deformation refers to the shape changes of the wing structure under aerodynamic loads during flight. It is an important component of component status data in flight operation data, monitored and recorded in real time by fiber optic sensors as deformation data, reflecting the structural safety status of the aircraft. Engine anomalies refer to the state where engine operating parameters deviate from the normal range. It is a core monitoring content of component status data in flight operation data; abnormal data collected by vibration sensors can directly trigger risk assessment warnings. Sudden gusts refer to sudden and strong changes in airflow in the atmosphere. They are key meteorological elements in environmental parameter data, collected and recorded in real time by wind speed sensors, directly affecting flight safety assessments and fault mode identification. Optical signal data refers to laser signals collected through fiber optic sensing networks deployed on the surface of mobile devices. Their optical properties, such as phase and intensity, change measurably with structural deformation, and are used to invert the microscopic deformation of the device surface.

[0068] In this embodiment, optical signal data is first collected in real time through a fiber optic sensor network installed on key parts of the mobile device. Simultaneously, component status data (such as vibration sensor readings) and environmental parameter data (such as meteorological sensor data) are obtained from the airborne monitoring system. All data is recorded with a unified timestamp to ensure temporal consistency in subsequent analysis.

[0069] For example, during flight, a mobile device uses fiber optic sensors to monitor wing deformation in real time and collect data on changes in light phase; simultaneously, engine vibration sensors record vibration amplitude data, and weather sensors record current wind speed and temperature data. All data is stored synchronously with millisecond-level timestamps to form the original dataset.

[0070] Step 102: Based on the optical signal data and the structural deformation field of the mobile device, a strain thermogram of the structural surface is formed.

[0071] In step 102, the structural deformation field represents the deformation distribution of the mobile device structure under stress, determined by a pre-defined fiber optic sensing network topology. The strain thermogram visually displays the strain intensity distribution in different regions of the structural surface using color gradients.

[0072] In this embodiment, the phase offset of the optical signal data is analyzed, and combined with the spatial location information of the fiber optic sensor, the local deformation corresponding to each sensing point is calculated. Based on the difference in deformation between adjacent sensing points, the strain intensity is calculated, and finally a thermal map covering the entire surface of the structure is generated, where the color depth represents the strain magnitude.

[0073] For example, based on the phase offset data collected by the fiber optic sensor, the deformation of three adjacent sensing points in a certain area of ​​the wing is calculated to be 0.12 mm, 0.15 mm and 0.10 mm, respectively. The local strain is calculated using the strain formula ε=ΔL / L (where ΔL is the difference in deformation and L is the distance between sensing points), and finally a thermal map showing the strain concentration at the root of the left wing is generated.

[0074] Step 103: Perform spatiotemporal alignment and fusion of the strain thermogram, the component state data, and the environmental parameter data, and extract spatiotemporal features from the fusion result through an enhanced attention mechanism.

[0075] In step 103, spatiotemporal alignment fusion means uniformly calibrating data from different sources in time and space to ensure data correlation.

[0076] In this embodiment, heatmaps, component status data, and environmental parameter data are sliced ​​at a uniform time and their location information is matched within a three-dimensional spatial grid. Dynamic weights are assigned to multi-source data within the same grid cell (e.g., strain data has a higher weight, environmental data has a lower weight), and a fusion tensor is generated through weighted fusion. Finally, an attention mechanism is used to enhance key spatiotemporal features and suppress redundant information.

[0077] For example, at time 30, the strain value in a certain grid cell on the left wing is 0.15 radians, the vibration value is 0.17 millimeters, and the wind speed is 25 meters per second. After weighted fusion with weights of 0.6, 0.3, and 0.1, the eigenvalue of this cell is 0.6 × 0.15 + 0.3 × 0.17 + 0.1 × 25 = 3.091. After attention-weighted fusion of the entire aircraft's grid data, the eigenvalue of the high-strain region on the left wing is improved.

[0078] Step 104: Perform kernel fuzzy C-means clustering on the spatiotemporal features to identify potential fault mode combinations, and determine the flight risk assessment results based on the potential fault mode combinations.

[0079] In step 104, kernel fuzzy C-means clustering maps the data to a high-dimensional space for fuzzy grouping, resolving the nonlinear separability problem. Potential fault mode combinations represent the set of typical fault features extracted after clustering. The flight risk assessment result refers to the quantitative safety level output after matching the potential fault mode combinations identified by kernel fuzzy C-means clustering with a preset risk level comparison table. Its specific meaning includes: 1) Risk level classification (e.g., level one to five, with higher values ​​indicating higher risk), reflecting the degree of matching between the current operating status and typical fault modes; 2) Fault type localization (e.g., structural deformation exceeding limits and strong turbulent coupling), clarifying the source of risk; 3) Early warning recommendations (e.g., requiring immediate repair or continuous monitoring), providing a basis for operation and maintenance decisions. This result is generated by mapping and comparing cluster feature combinations with historical accident data, encompassing both a comprehensive assessment of the current state and a prediction of fault development trends.

[0080] In this embodiment, spatiotemporal features are mapped to a high-dimensional kernel space, several cluster centers are initialized, and the center positions are adjusted through iterative optimization. The membership degree between each feature vector and the center is calculated to generate a correlation strength matrix, and finally, clusters are divided according to the maximum membership degree. The core feature combinations of each group are extracted and matched with a preset risk level table to output the evaluation results.

[0081] For example, a cluster analysis identified "vibration 0.2 mm + strain 0.16 radians + wind speed 28 m / s" as a fault mode, which was determined to be a level three risk according to the risk table, triggering an early warning.

[0082] This method achieves accurate identification and risk assessment of complex faults under complex working conditions through dynamic fusion of multi-source data and intelligent clustering. It solves the problems of insufficient sensitivity to sudden anomalies and high misjudgment rate of traditional methods, and improves the reliability of safety monitoring.

[0083] To address the issue of insufficient accuracy in identifying composite fault modes in flight operation data, in some embodiments, step 104: performing kernel fuzzy C-means clustering on the spatiotemporal features to identify potential fault mode combinations, and determining the flight risk assessment result based on the potential fault mode combinations, includes:

[0084] Step 201: Map the spatiotemporal features to the kernel space to generate a feature vector set. The kernel space is an implicit feature space generated by kernel function transformation in kernel fuzzy C-means clustering.

[0085] In step 201, the kernel space refers to the high-dimensional feature space to which the original feature data is mapped through mathematical transformation. This transformation can better distinguish complex data patterns. The feature vector set is a high-dimensional data set formed after the spatiotemporal features are transformed by the kernel function.

[0086] In this embodiment, a suitable kernel function is first selected to perform a nonlinear transformation on the spatiotemporal features, mapping the feature data that was originally difficult to separate linearly to a higher-dimensional space. In this new space, similar features will automatically cluster together, creating favorable conditions for subsequent clustering analysis.

[0087] Step 202: Initialize the cluster centers of the feature vector set, and form a cluster group set based on the cluster centers and the feature vector set.

[0088] In step 202, the cluster centroid refers to the central position representing each feature group in the high-dimensional kernel space. The cluster set is a set of feature vectors grouped by similarity using an optimization algorithm.

[0089] In this embodiment of the application, several initial center point positions are first randomly set, and then the positions of these center points are continuously adjusted through iterative calculations so that each feature vector can find the center point most similar to it, and finally several feature groups are formed, with the features in each group having high similarity.

[0090] Step 203: Extract the core feature combination of each cluster group in the cluster group set as a potential fault mode combination.

[0091] In step 203, the core feature combination consists of the coupling states of three key features: 1) structural strain features (such as the deformation range of 0.15-0.18 radians on the wing surface); 2) component state features (such as the vibration amplitude of the engine at 0.20-0.22 mm); and 3) environmental features (such as a gust wind speed of 28-30 m / s). For example, when cluster analysis identifies the feature combination of "wing deformation of 0.16 radians + engine vibration of 0.21 mm + encountering a gust wind of 29 m / s," it constitutes a typical fault mode combination reflecting the coupling of aerodynamic loads and mechanical vibrations.

[0092] In this embodiment of the application, statistical analysis is performed on all feature vectors within each cluster group to find the concentrated distribution interval of feature values ​​for each group. Combining these intervals forms a representative fault feature pattern.

[0093] Step 204: Based on the preset risk level comparison table, map the potential failure mode combinations into flight risk assessment results.

[0094] In step 204, the preset risk level comparison table is a pre-established table that corresponds to the combination of fault characteristics and the degree of risk.

[0095] In this embodiment of the application, the identified feature combination is compared with a pre-established fault case library to find the closest historical fault mode, thereby determining the risk level of the current equipment operating status.

[0096] Here is a specific example:

[0097] Following the scenario where the mobile device detected concentrated strain at the root of the left wing and abnormal engine vibration during flight, after the system acquired the characteristic data at 30 seconds (left wing strain 0.15 radians, vibration 0.17 millimeters, wind speed 25 m / s), it first applied a Gaussian kernel function. The features are mapped to a high-dimensional space, where x represents the feature vector and y represents the kernel center point. σ, set to 0.5, controls the mapping range. The kernel space coordinates of the feature vector at this moment are calculated to be [0.82, 0.15, 0.03]. The coordinates of the three cluster center points are initialized to [0.80, 0.20, 0.00], [0.10, 0.90, 0.00], and [0.30, 0.30, 0.40], and updated to [0.85, 0.12, 0.03], [0.15, 0.80, 0.05], and [0.25, 0.25, 0.50] after five iterations. The membership degree of the feature vector to each center point is calculated using the formula... Where d is the kernel spatial distance, the correlation strengths between this vector and the three centers are 0.92, 0.08 and 0.00 respectively, so it is assigned to the first cluster group; finally, the core feature range extracted from this group is a combination of strain 0.14-0.18 radians, vibration 0.16-0.20 mm and wind speed 24-26 m / s. After matching with the risk level table, it is confirmed as a level 2 risk state that needs to be monitored. The system then issues a graded early warning signal to prompt the operator to strengthen the monitoring of the area.

[0098] In this embodiment, the synergistic effect of kernel space transformation and fuzzy clustering enables accurate capture of complex fault characteristics and intelligent determination of risk levels, effectively improving the ability to identify safety hazards in complex operating environments and providing a reliable basis for equipment maintenance decisions.

[0099] To address the issues of insufficient accuracy in optimizing cluster center locations and feature grouping during flight data clustering analysis, in some embodiments, step 202: forming a cluster set based on the cluster center points and the feature vector set includes:

[0100] Step 301: Use a genetic algorithm to iteratively optimize the position of the cluster center point.

[0101] In step 301, the genetic algorithm operation is an optimization method that simulates the natural evolution process, and gradually improves the position of the cluster center point through operations such as selection, crossover and mutation.

[0102] In this embodiment, several sets of center point location schemes are first randomly generated, the clustering effect index corresponding to each scheme is calculated, the scheme with the best effect is retained as the parent, and new child schemes are generated through position crossover and small perturbations. After multiple iterations, the optimal center point distribution is obtained.

[0103] Step 302: Based on the optimized position, generate the association strength matrix using a preset membership function.

[0104] In step 302, the membership function is used to calculate the degree of association between the eigenvectors and each centroid. The association strength matrix is ​​a two-dimensional table that records the association strength between all eigenvectors and each centroid.

[0105] In this embodiment of the application, the distance from each feature vector to each center point is calculated based on the optimized center point position. The distance is then converted into a correlation strength value between 0 and 1 using a preset conversion function, and finally a strength matrix is ​​generated with rows representing feature vectors and columns representing center points.

[0106] Step 303: Determine the final clustering result of each feature vector in the feature vector set based on the maximum index of the row vector of the association strength matrix.

[0107] In step 303, the row vector maximum value index refers to the column number of the maximum value in each row of the association strength matrix. The final clustering assignment result is the centroid number assigned to each feature vector.

[0108] In this embodiment of the application, each row of the association strength matrix is ​​scanned, the cell with the largest value in that row is found, and the column number corresponding to that cell is used as the final category of the feature vector of that row.

[0109] Step 304: Based on the final clustering result, group all feature vectors according to their respective cluster centers to form cluster groups corresponding to each cluster center, and integrate all cluster groups to form a cluster group set.

[0110] In step 304, a cluster group refers to the set of all feature vectors belonging to the same centroid.

[0111] In this embodiment of the application, all feature vectors are classified into corresponding centroids according to the attribution results. The feature vectors under each centroid form an independent group, and these groups together constitute a complete set of cluster groups.

[0112] Here is a specific example:

[0113] Based on the kernel space feature vector [0.82, 0.15, 0.03] obtained by the aforementioned mobile device at 30 seconds, the system first uses a genetic algorithm to optimize the initial center point positions. The three initially set center points [0.80, 0.20, 0.00], [0.10, 0.90, 0.00], and [0.30, 0.30, 0.40] are updated to [0.85, 0.12, 0.03], [0.15, 0.80, 0.05], and [0.25, 0.25, 0.50] after iterative optimization. In each iteration, the cluster density index J=Σμ²d² corresponding to each center point scheme is calculated for optimization, where μ represents the membership degree and d represents the distance from the feature vector to the center point. Then, the correlation strength between the feature vector and the optimized center point is calculated using the formula... The calculated association values ​​with the three centers are 0.92, 0.08, and 0.00, respectively, where d is calculated using ||xy||, where x represents the coordinates of the feature vector and y represents the coordinates of the center point. Based on the maximum value of 0.92, the feature vector is assigned to the cluster group corresponding to the first center point. Finally, the partitioning results of all 200 feature vectors are integrated to form three cluster groups.

[0114] In this embodiment, by intelligently optimizing the center point position and accurately calculating the feature correlation, efficient grouping of flight data is achieved, ensuring the close aggregation of similar features and the clear separation of different types of features, thus providing a reliable data foundation for subsequent fault mode identification.

[0115] To address the issue of insufficient accuracy in quantifying the association between feature vectors and cluster centers in flight data clustering analysis, in some embodiments, step 302: generating an association strength matrix based on the optimized position using a preset membership function includes:

[0116] Step 401: Based on the optimized position, calculate the similarity metric from each feature vector in the feature vector set to each cluster center point in the kernel space.

[0117] In step 401, the similarity metric refers to the degree of proximity between the feature vector and the cluster center in the kernel space. The larger the value, the more similar the features are.

[0118] In this embodiment, the optimized cluster center point coordinates and kernel space coordinates of the feature vector set are first obtained. Then, the distance from each feature vector to each center point is calculated. Finally, the distance is converted into a similarity value using a preset similarity calculation formula.

[0119] Step 402: Combine all similarity metrics corresponding to each feature vector into a similarity vector.

[0120] In step 402, the similarity vector refers to the set of similarity values ​​between a single feature vector and all cluster centers, and its dimension is equal to the number of cluster centers.

[0121] In the embodiments of this application, for each feature vector, its similarity values ​​with each center are arranged in a fixed order of center points to form an ordered numerical sequence, which is the similarity vector of the feature vector.

[0122] Step 403: Input the similarity vector into the membership function and output the association strength value of the feature vector with each cluster center point.

[0123] In step 403, the association strength value refers to the probability value of the feature vector belonging to a certain cluster center after the membership function standardization process, which is between 0 and 1.

[0124] In this embodiment, the similarity vector is input into a preset transformation function to normalize the original similarity, so that the sum of the association strength of each feature vector with all center points is 1, thereby obtaining a probabilistic association strength distribution.

[0125] Step 404: Aggregate the association strength values ​​of all feature vectors and construct the association strength matrix between each feature vector and each cluster center point.

[0126] In this embodiment, the correlation strength value of each vector is sequentially filled into the corresponding row of the matrix according to the fixed order of the feature vectors, and finally a complete matrix is ​​formed with the number of rows equal to the number of feature vectors and the number of columns equal to the number of center points.

[0127] Here is a specific example:

[0128] Based on the kernel space feature vector [0.82, 0.15, 0.03] obtained by the aforementioned mobile device at 30 seconds and the optimized positions of the three center points [0.85, 0.12, 0.03], [0.15, 0.80, 0.05], and [0.25, 0.25, 0.50], the Euclidean distance from this feature vector to each center point is first calculated, where the distance... , , Then, these distance values ​​are converted into similarity metrics using the similarity calculation formula S=1 / (1+d²), yielding 0.983, 0.527, and 0.740. These three similarity values ​​are then combined to form a similarity vector [0.983, 0.527, 0.740], which is then normalized using the membership function μ=S / ∑S, where ∑S=0.983+0.527+0.740=2.250. Finally, the association strength between this feature vector and the three centroids is obtained. The values ​​are 0.437, 0.234, and 0.329, respectively. After performing the same calculation process on all 200 feature vectors, the system constructs a 200-row, 3-column association strength matrix. Each row of data represents the degree of association between the corresponding feature vector and the three centroids. For example, the first row [0.437, 0.234, 0.329] represents the probability distribution of the first feature vector and the three centroids. This matrix provides a quantitative basis for accurately classifying feature vectors into corresponding cluster groups.

[0129] In this embodiment of the application, a scientific correlation strength quantification system was established by accurately calculating feature similarity and standardizing transformation, which ensures the objectivity and accuracy of feature vector clustering and grouping, and provides reliable data support for flight safety risk assessment.

[0130] To address the issue of insufficient accuracy in spatiotemporal feature extraction during multi-source flight data fusion, some embodiments include step 103: performing spatiotemporal alignment fusion of the strain heatmap, the component state data, and the environmental parameter data, and extracting spatiotemporal features from the fusion result using an enhanced attention mechanism, including:

[0131] Step 501: Perform independent time slices on the strain thermogram, the component status data, and the environmental parameter data according to a unified timestamp to obtain three types of time slice groups at the same time.

[0132] In step 501, a time slice group refers to an instantaneous data segment extracted from a continuous data stream at the same time point. The three types of time slice groups correspond to the instantaneous states of strain, component, and environmental data, respectively.

[0133] In this embodiment of the application, a unified time reference axis is first established, and then data segments at the same time are extracted synchronously from the three types of raw data according to the set time interval to ensure that all data are strictly aligned in the time dimension.

[0134] Step 502: Based on the structural space model of the mobile device, map the three types of time slice groups to a preset three-dimensional spatial grid coordinate system.

[0135] In step 502, the structural space model of the mobile device is a pre-established three-dimensional digital model of the device, containing the geometric dimensions, material properties, and sensor placement information of all key components. This model is used to uniformly map data from different sources to the same physical space coordinate system, ensuring that strain, vibration, and other data accurately correspond to the specific locations on the structure. The three-dimensional spatial grid coordinate system is a three-dimensional grid system formed by dividing the mobile device structure at a certain resolution, with each grid cell corresponding to a specific location on the structure.

[0136] In this embodiment of the application, a grid coordinate system is pre-established based on the structural drawings of the device, and then each type of data is mapped to the corresponding grid cell according to its collection location information, so that data from different sources are accurately matched in spatial location.

[0137] Step 503: In the three-dimensional spatial grid coordinate system, the three types of time slice groups within the same spatial grid cell are dynamically weighted and fused to generate a fused grid tensor.

[0138] In step 503, the fused mesh tensor refers to the three-dimensional data block formed by fusing multi-source data according to weights in a unified spatiotemporal grid.

[0139] In this embodiment of the application, three types of data within each grid cell are analyzed, and fusion weights are dynamically assigned based on the importance and reliability of the data. Then, the weighted data are superimposed and integrated to form tensor data containing complete feature information.

[0140] Step 504: Perform a spatiotemporal attention weighting operation on the fused mesh tensor to generate spatiotemporal features.

[0141] In step 504, spatiotemporal attention weighting refers to the process of dynamically adjusting the importance of data based on its temporal and spatial variation characteristics. The spatiotemporal features are represented as a multidimensional vector sequence after attention weighting, with each vector containing a quantified value of risk features at a specific time and spatial location (e.g., a risk coefficient of 0.82 at the left wing leading edge at 30 seconds), while the fusion grid tensor is the unweighted original fusion data cube; the former highlights key risk signals, while the latter retains all original information. The difference between the two is that the former is a refined and risk-focused expression of the latter.

[0142] In this embodiment, the variation patterns of the fusion tensor in various dimensions are analyzed, higher weights are assigned to fluctuating regions, and lower weights are assigned to stable regions, ultimately extracting a spatiotemporal feature set that reflects key risk characteristics.

[0143] Here is a specific example:

[0144] Based on the flight data collected by the aforementioned mobile device at 30 seconds, the system first synchronously slices the strain heatmap, engine vibration data, and ambient wind speed data of the left wing root region according to a time window of 30.0-30.9 seconds, obtaining 10 sets of time-aligned data slices. The slice at 30.5 seconds includes a strain value of 0.15 radians, a vibration value of 0.17 millimeters, and a wind speed of 25 meters per second. According to the pre-stored wing 3D structural model, these data are mapped to a grid cell numbered G205, corresponding to a specific location on the left wing leading edge. During the fusion process, the system automatically assigns weights based on data importance, with strain data having a weight of 0.6... With vibration data of 0.3 and wind speed data of 0.1, the initial fusion value of the unit was obtained as 3.091 through weighted calculation of 0.6×0.15+0.3×0.17+0.1×25. Subsequently, attention-weighted analysis was performed, and it was found that the strain fluctuation in this region reached 0.12-0.18 radians during the period of 30.0-30.9 seconds. Therefore, an attention weight of 0.9 was assigned, and the final eigenvalue of the unit was 3.091×0.9=2.782. At the same time, the corresponding region on the right wing only received a weight of 0.3 due to the gradual change in data, and the eigenvalue was 1.953. After the system processed all grid units of the entire machine, a complete spatiotemporal feature dataset was formed.

[0145] In this embodiment, through precise spatiotemporal alignment and intelligent weight allocation, efficient fusion and key feature extraction of multi-source flight data are achieved, laying a data foundation for accurate identification of complex fault modes and improving the reliability of flight safety monitoring.

[0146] To address the issue of inaccurate key feature extraction after multi-source data fusion, in some embodiments, step 504: performing a spatiotemporal attention weighting operation on the fused mesh tensor to generate spatiotemporal features includes:

[0147] Step 601: Divide the fused mesh tensor into multiple time windows along the time dimension, and generate time attention weights based on the changing trend of mesh data within each time window.

[0148] In step 601, the trend of grid data change within the time window refers to the degree of data fluctuation within each time period. This is obtained by calculating the standard deviation of the data in each grid cell within that time period; a larger standard deviation indicates more severe fluctuations. Specifically, this is achieved by taking a data sequence of consecutive time points within the time window and calculating its statistical dispersion. The time attention weight refers to the weight value assigned based on the severity of data changes over time; the more volatile the time period, the higher the weight.

[0149] In this embodiment, the fused mesh tensor is first divided into several time segments with a fixed duration, then the fluctuation amplitude of the data in each segment is calculated, and finally, the corresponding time attention weights are assigned according to the fluctuation magnitude.

[0150] Step 602: In the spatial dimension, generate spatial attention weights based on the data variation amplitude between grid cells in the fused grid tensor.

[0151] In step 602, the magnitude of data variation between grid cells refers to the degree of difference between data in adjacent spatial locations. This is obtained by calculating the gradient values ​​of adjacent grid cell data; a larger gradient value indicates greater spatial variation. Specifically, it is calculated by averaging the absolute values ​​of the differences between the data in the central grid cell and the data in the surrounding cells. Spatial attention weight refers to the weight values ​​assigned based on the degree of difference in the spatial distribution of data; regions with more significant differences have higher weights.

[0152] In this embodiment, the data change gradient between adjacent grid cells is analyzed, and regions with drastic changes are assigned higher spatial attention weights, while regions with stable changes are assigned lower weights.

[0153] Step 603: Decompose the fused mesh tensor into multiple feature channels according to the data source type, and generate channel attention weights based on the covariance between the feature channels.

[0154] In step 603, the covariance between feature channels refers to the strength of the correlation between different data sources. It is obtained by calculating the covariance matrix of the data sequences of each channel. The larger the absolute value of the covariance, the stronger the correlation. Specifically, it is calculated by using the formula cov(X,Y)=E[(X-μx)(Y-μy)]. Channel attention weight refers to the weight value assigned according to the strength of the correlation between different data sources. The stronger the correlation, the higher the weight of the channel.

[0155] In this embodiment, the covariance matrix between each data channel is calculated, and corresponding channel attention weights are assigned according to the correlation between the channels.

[0156] Step 604: Perform tensor product operation on the temporal attention weight, the spatial attention weight, and the channel attention weight to construct a three-dimensional attention weight tensor.

[0157] In step 604, the three-dimensional attention weight tensor is a weight distribution cube formed by comprehensively calculating the weights of the three dimensions of time, space and channel.

[0158] In this embodiment of the application, the three-dimensional weight matrices are subjected to tensor product operation to generate a three-dimensional weight tensor that is completely consistent with the dimensions of the fused mesh tensor.

[0159] Step 605: Multiply the three-dimensional attention weight tensor element-wise with the fused mesh tensor to output the spatiotemporal features.

[0160] In step 605, element-wise multiplication refers to the operation of multiplying the values ​​of the weight tensor and the fusion tensor at corresponding positions pairwise.

[0161] In this embodiment of the application, the three-dimensional attention weight tensor and the fusion mesh tensor are multiplied by the numerical values ​​corresponding to their positions to finally obtain spatiotemporal feature data that strengthens key features and weakens secondary features.

[0162] Here is a specific example:

[0163] Based on the fused mesh data generated by the aforementioned mobile device during the 30.0-30.9 second period, the system first divides this period into three time windows (30.0-30.3 seconds, 30.3-30.6 seconds, and 30.6-30.9 seconds). The standard deviations of the strain values ​​of the left-wing G205 mesh element in each window are calculated to be 0.02, 0.03, and 0.01 radians, respectively. Time attention weights are allocated according to the standard deviation ratio: 0.4, 0.6, and 0.2. In the spatial dimension, the average strain difference between the G205 element and its three adjacent elements is calculated to be 0.04 radians, and the vibration difference is 0.03 millimeters. A spatial attention weight of 0.7 is derived by combining these values. The strain, vibration, and wind speed data are then processed according to the covariance formula. The strain-vibration covariance was calculated to be 0.85, strain-wind speed to be 0.62, and vibration-wind speed to be 0.45. Based on these, channel weights of 0.5, 0.3, and 0.2 were assigned. Tensor product operation was performed on the weights of the three dimensions (0.6 (time), 0.7 (space), and 0.5 (channel)) to obtain the composite weight of the G205 element in the 30.3-30.6 second window: 0.6 × 0.7 × 0.5 = 0.21. Multiplying this weight by the original fusion value of 3.091 yielded the final eigenvalue of 0.65. Meanwhile, the corresponding region on the right wing had a final eigenvalue of only 0.19 due to the lower weights of each dimension (0.3, 0.2, 0.3). The system enhanced the high-risk characteristics of the left wing leading edge through this differentiated weighting.

[0164] In this embodiment, a three-dimensional attention weighting mechanism is used to achieve intelligent focusing on key features of flight data, effectively improving the sensitivity of identifying composite fault features and providing more accurate data support for flight safety early warning.

[0165] To address the issue of insufficient visualization accuracy of strain distribution in structural deformation monitoring, in some embodiments, step 102: forming a strain thermogram of the structural surface based on the optical signal data and the structural deformation field of the mobile device, includes:

[0166] Step 701: Analyze the phase offset distribution from the optical signal data.

[0167] In step 701, the phase offset distribution refers to the spatial distribution of the phase change of the optical signal collected by the fiber optic sensor, reflecting the degree of deformation of each part of the structure.

[0168] In this embodiment, the original optical signal is first demodulated to extract phase change data from each sensing point, forming a phase change distribution map covering the entire monitoring area. The specific implementation process is as follows: After acquiring the original optical signal through each sensing point in the fiber optic sensor network, the original optical signal is input into a phase demodulation device. The device separates the interference optical signal component carrying deformation information, performs phase calculation on the interference optical signal component, and outputs the real-time phase offset value corresponding to each sensing point. Based on the spatial layout coordinates of the sensing points, the phase offset value is mapped to the corresponding area on the wing surface. Example: Taking aircraft wing health monitoring as an example, a fiber optic sensor array is deployed on the wing surface in a 20cm × 20cm grid. When the aircraft encounters turbulence: each fiber optic sensing point acquires the original optical signal in real time and transmits it to the onboard demodulation unit; the demodulation unit performs interference fringe analysis on the optical signal at the third grid point on the left wing and extracts a phase offset of 0.15 rad.

[0169] Step 702: Based on the pre-set fiber optic sensor network topology in the structural deformation field of the mobile device, bind the phase offset to the corresponding spatial location point.

[0170] In step 702, the fiber optic sensor network topology refers to the spatial description of the placement and connection method of the fiber optic sensors on the structural surface.

[0171] In this embodiment of the application, according to the pre-stored optical fiber deployment scheme, a correspondence is established between each phase change data point and the specific location coordinates of the structural surface to ensure accurate matching between data and spatial location.

[0172] Step 703: Based on each spatial location point and its corresponding phase offset, calculate the local deformation using a preset phase and deformation transformation model.

[0173] In step 703, the phase and deformation conversion model is essentially a physical relationship model established based on the fiber optic grating sensing principle. For example, the standard photoelastic effect formula is used: deformation = phase offset × fiber wavelength / (2π × photoelastic coefficient), where the photoelastic coefficient is determined by the fiber material properties, realizing the quantitative conversion of optical signal phase change to physical deformation.

[0174] In this embodiment of the application, for each located phase data point, the actual deformation at that location is calculated using a preset formula based on the physical properties of the material and the optical fiber parameters.

[0175] Step 704: Analyze the differences in local deformation between adjacent spatial locations to determine the strain intensity distribution on the structural surface.

[0176] In step 704, "adjacent" refers to sensing points in the fiber optic sensing network whose physical distance does not exceed a preset threshold. Typically, at least three adjacent points (front / back / left or right) are required to calculate a reliable strain gradient and ensure spatial continuity. "Difference information" refers to the rate of change of deformation between adjacent sensing points, obtained by calculating the ratio of the deformation difference to the distance between sensing points. It reflects the severity of local tension / compression in the structure. For example, the difference between a deformation of 0.1 mm at point A and a deformation of 0.15 mm at adjacent point B is 0.05 mm / 10 cm = 500 microstrain. "Strain intensity distribution" is a quantitative description of the severity of deformation in different regions of the structural surface.

[0177] In this embodiment, the deformation difference between adjacent measurement points is calculated, and the strain intensity value of each local area is obtained by combining the point spacing data to form a complete strain distribution field.

[0178] Step 705: Perform data visualization transformation on the strain intensity distribution according to the preset color gradient rules to generate a strain thermogram of the structural surface.

[0179] In step 705, the color gradient rule refers to a visualization scheme that maps strain values ​​to different colors.

[0180] In this embodiment of the application, the strain intensity value of each point is converted into the corresponding color code according to the preset color mapping table, and finally a heat map that intuitively displays the strain distribution is generated.

[0181] Here is a specific example:

[0182] Based on the aforementioned scenario of monitoring the left wing during flight using a mobile device, the fiber optic sensor network collected optical signal phase offset data of 0.12 radians, 0.15 radians, and 0.10 radians from three adjacent sensing points (numbered P1, P2, and P3, spaced 0.1 meters apart). According to a pre-defined fiber optic deployment topology, the coordinates of these three points corresponding to the leading edge of the left wing were determined as (X12, Y05), (X13, Y05), and (X14, Y05). The deformation was calculated using the phase-deformation conversion model ΔL = λ·Δφ / 2πn, where λ is the optical wavelength (15 nm). With a fiber refractive index of 1.45 and a phase offset of 50 nm, the deformation values ​​at three points are 0.12 mm, 0.15 mm, and 0.10 mm, respectively. The strain between P1 and P2 is calculated as ε = (0.15 - 0.12) / 0.1 = 0.3, and the strain between P2 and P3 is calculated as ε = (0.15 - 0.10) / 0.1 = 0.5. According to the color mapping rule, the strain value of 0.3 corresponds to yellow and 0.5 corresponds to red. In the final generated thermal map, the P2-P3 area is displayed as a red warning, and the P1-P2 area is displayed as a yellow prompt.

[0183] In this embodiment of the application, the visual conversion of fiber optic sensing data into strain distribution enables an intuitive presentation of the structural deformation state, providing a reliable basis for deformation assessment for flight safety monitoring.

[0184] Figure 2 This application provides a schematic diagram of the structure of a system for mining spatiotemporal features and assessing risks from flight operation data, as shown in the embodiments of this application. Figure 2 As shown, the system includes:

[0185] The acquisition module 21 is used to acquire flight operation data and optical signal data of the mobile device. The flight operation data includes component status data and environmental parameter data.

[0186] The forming module 22 is used to form a strain thermogram of the structural surface based on the optical signal data and the structural deformation field of the mobile device.

[0187] Extraction module 23 is used to perform spatiotemporal alignment and fusion of the strain thermogram, the component state data and the environmental parameter data, and extract spatiotemporal features from the fusion result through an enhanced attention mechanism.

[0188] The identification module 24 is used to perform kernel fuzzy C-means clustering on the spatiotemporal features to identify potential fault mode combinations and determine the flight risk assessment result based on the potential fault mode combinations.

[0189] Figure 2 The aforementioned system for spatiotemporal feature mining and risk assessment of flight operation data can perform... Figure 1The implementation principle and technical effects of the spatiotemporal feature mining and risk assessment method for flight operation data described in the illustrated embodiment will not be repeated here. The specific methods of operation of each module and unit in the spatiotemporal feature mining and risk assessment system for flight operation data in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0190] In one possible design, Figure 2 The spatiotemporal feature mining and risk assessment system for flight operation data shown in the embodiment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0191] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0192] The processing component 32 is used to perform the above. Figure 1 The embodiment describes a method for spatiotemporal feature mining and risk assessment of flight operation data.

[0193] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0194] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented from 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.

[0195] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0196] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0197] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0198] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0199] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown illustrates a method for spatiotemporal feature mining and risk assessment of flight operation data.

[0200] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0201] 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.

[0202] 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.

[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for spatiotemporal feature mining and risk assessment of flight operation data, characterized in that, include: Acquire flight operation data and optical signal data of the mobile device, wherein the flight operation data includes component status data and environmental parameter data; Based on the optical signal data and the structural deformation field of the mobile device, a strain thermogram of the structural surface is formed. The strain thermogram, the component state data, and the environmental parameter data are spatiotemporally aligned and fused, and spatiotemporal features are extracted from the fusion result through an enhanced attention mechanism. Kernel fuzzy C-means clustering is performed on the spatiotemporal features to identify potential failure mode combinations, and flight risk assessment results are determined based on the potential failure mode combinations. The step of performing kernel fuzzy C-means clustering on the spatiotemporal features to identify potential failure mode combinations, and determining the flight risk assessment result based on the potential failure mode combinations, includes: The spatiotemporal features are mapped to the kernel space to generate a feature vector set. The kernel space is an implicit feature space generated by kernel function transformation in kernel fuzzy C-means clustering. Initialize the cluster centers of the feature vector set, and form a set of cluster groups based on the cluster centers and the feature vector set; Extract the core feature combinations of each cluster in the cluster set as potential fault mode combinations; Based on a preset risk level comparison table, the potential failure mode combinations are mapped to flight risk assessment results; The step of spatiotemporally aligning and fusing the strain thermogram, the component state data, and the environmental parameter data, and extracting spatiotemporal features from the fusion result using an enhanced attention mechanism, includes: The strain thermogram, the component status data, and the environmental parameter data are each independently sliced ​​according to a unified timestamp to obtain three types of time slice groups at the same time. Based on the structural space model of the mobile device, the three types of time slice groups are uniformly mapped to a preset three-dimensional spatial grid coordinate system; In the three-dimensional spatial grid coordinate system, three types of time slice groups within the same spatial grid cell are dynamically weighted and fused to generate a fused grid tensor; Spatiotemporal attention weighting is performed on the fused mesh tensor to generate spatiotemporal features.

2. The method for spatiotemporal feature mining and risk assessment of flight operation data according to claim 1, characterized in that, The process of forming a cluster group set based on the cluster centroids and the feature vector set includes: The positions of the cluster centers are iteratively optimized using a genetic algorithm. Based on the optimized position, a correlation strength matrix is ​​generated using a preset membership function; Based on the index of the maximum value of the row vector in the association strength matrix, the final clustering result of each feature vector in the feature vector set is determined; Based on the final clustering results, all feature vectors are grouped according to their respective cluster centers to form cluster groups corresponding to each cluster center. All cluster groups are then integrated to form a cluster group set.

3. The method for spatiotemporal feature mining and risk assessment of flight operation data according to claim 2, characterized in that, The step of generating a correlation strength matrix based on the optimized position using a preset membership function includes: Based on the optimized position, calculate the similarity metric from each feature vector in the feature vector set to each cluster center point in the kernel space; For each feature vector, all similarity metrics are combined to form a similarity vector. Input the similarity vector into the membership function to output the association strength value of the feature vector with each cluster center point; Aggregate the association strength values ​​of all feature vectors to construct an association strength matrix between each feature vector and each cluster centroid.

4. The method for spatiotemporal feature mining and risk assessment of flight operation data according to claim 1, characterized in that, The step of performing a spatiotemporal attention weighting operation on the fused mesh tensor to generate spatiotemporal features includes: The fused mesh tensor is divided into multiple time windows along the time dimension, and time attention weights are generated based on the changing trend of mesh data within each time window. In the spatial dimension, spatial attention weights are generated based on the magnitude of data variation between grid cells in the fused grid tensor. The fused mesh tensor is decomposed into multiple feature channels according to the data source type, and channel attention weights are generated based on the covariance between the feature channels. The temporal attention weights, spatial attention weights, and channel attention weights are multiplied by tensors to construct a three-dimensional attention weight tensor. The spatiotemporal features are output by multiplying the three-dimensional attention weight tensor element-wise with the fused mesh tensor.

5. The method for spatiotemporal feature mining and risk assessment of flight operation data according to claim 1, characterized in that, The step of forming a strain thermogram of the structural surface based on the optical signal data and the structural deformation field of the mobile device includes: The phase offset distribution is analyzed from the optical signal data; Based on the pre-set fiber optic sensor network topology in the structural deformation field of the mobile device, the phase offset is bound to the corresponding spatial location point; Based on each spatial location point and its corresponding phase offset, the local deformation is calculated using a preset phase and deformation transformation model. Analyze the differences in local deformation between adjacent spatial locations to determine the strain intensity distribution on the structural surface; The strain intensity distribution is transformed into a data visualization according to a preset color gradient rule to generate a strain thermogram of the structural surface.

6. A system for spatiotemporal feature mining and risk assessment of flight operation data, characterized in that, include: The acquisition module is used to acquire flight operation data and optical signal data of the mobile device, wherein the flight operation data includes component status data and environmental parameter data; A forming module is used to form a strain thermogram of the structural surface based on the optical signal data and the structural deformation field of the mobile device. The extraction module is used to perform spatiotemporal alignment and fusion of the strain thermogram, the component state data and the environmental parameter data, and extract spatiotemporal features from the fusion result through an enhanced attention mechanism; The identification module is used to perform kernel fuzzy C-means clustering on the spatiotemporal features to identify potential fault mode combinations, and determine the flight risk assessment result based on the potential fault mode combinations. The step of performing kernel fuzzy C-means clustering on the spatiotemporal features to identify potential failure mode combinations, and determining the flight risk assessment result based on the potential failure mode combinations, includes: The spatiotemporal features are mapped to the kernel space to generate a feature vector set. The kernel space is an implicit feature space generated by kernel function transformation in kernel fuzzy C-means clustering. Initialize the cluster centers of the feature vector set, and form a set of cluster groups based on the cluster centers and the feature vector set; Extract the core feature combinations of each cluster in the cluster set as potential fault mode combinations; Based on a preset risk level comparison table, the potential failure mode combinations are mapped to flight risk assessment results; The step of spatiotemporally aligning and fusing the strain thermogram, the component state data, and the environmental parameter data, and extracting spatiotemporal features from the fusion result using an enhanced attention mechanism, includes: The strain thermogram, the component status data, and the environmental parameter data are each independently sliced ​​according to a unified timestamp to obtain three types of time slice groups at the same time. Based on the structural space model of the mobile device, the three types of time slice groups are uniformly mapped to a preset three-dimensional spatial grid coordinate system; In the three-dimensional spatial grid coordinate system, three types of time slice groups within the same spatial grid cell are dynamically weighted and fused to generate a fused grid tensor; Spatiotemporal attention weighting is performed on the fused mesh tensor to generate spatiotemporal features.

7. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a method for spatiotemporal feature mining and risk assessment of flight operation data as described in any one of claims 1 to 5.

8. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements a method for spatiotemporal feature mining and risk assessment of flight operation data as described in any one of claims 1 to 5.

Images