An aviation equipment landing gear structure load on-board monitoring method and system based on deep learning model edge computing

Abstract

The application provides an airborne monitoring method and system for structural load of an aircraft landing gear based on edge computing of a deep learning model, and belongs to the technical field of structural load monitoring of an aircraft.The method provided by the application constructs a structural load prediction model based on historical load spectrum test data, determines an input feature set of three-way structural load through multi-dimensional feature screening, constructs a segmented prediction model in combination with the phased characteristics of the landing process of the landing gear, and deploys the model on an airborne edge computing unit to realize real-time prediction and output of the three-way structural load of the landing gear.The application solves the technical problems in the prior art that the structural load of the landing gear is difficult to realize real-time prediction on an aircraft, the selection of input features lacks systematicness, and the load prediction accuracy of different landing stages is difficult to balance.

Description

Technical Field

[0001] This invention belongs to the field of aviation equipment structural load monitoring technology, and particularly relates to an airborne monitoring method and system for aviation equipment landing gear structural load based on edge computing of a deep learning model. Background Technology

[0002] During takeoff, landing, and ground taxiing, the landing gear, as a critical load-bearing component, is subjected to complex impact loads and multi-directional alternating loads. Especially during the impact phase of landing and the braking phase of the taxiing run, the landing gear structure is subjected to significant vertical, longitudinal, and lateral load coupling effects, which directly affect the structural safety margin and operational reliability.

[0003] In current engineering practice, the main methods for obtaining landing gear structural loads include ground load spectrum tests, finite element simulation analysis, and a small number of strain or overload measurements at key locations. While ground load spectrum tests can provide relatively accurate load data, their results are usually used for offline analysis and are difficult to implement for real-time airborne monitoring. The finite element method relies on idealized models and boundary conditions, making it difficult to reflect the complex dynamic response under actual flight conditions. Single-sensor monitoring methods cannot fully characterize the coupling relationship of multi-directional loads on the landing gear.

[0004] Furthermore, with the intelligent development of aviation equipment and the gradual improvement of airborne real-time data processing capabilities, under the conditions of limited computing power, limited power consumption, and strict environmental adaptability requirements, the following problems still exist in how to deploy complex structural load prediction models on airborne edge computing units to achieve real-time load prediction: (1) There are complex nonlinear relationships between the landing gear structural load and flight parameters, compression stroke signal, overload signal and strain signal, and traditional linear methods are difficult to achieve high-precision prediction; (2) There are many types of input parameters, and the sensitivity of loads in different directions to input features varies. If effective feature selection is not performed, it will lead to increased model complexity and reduced inference efficiency. (3) The landing gear landing process has obvious phased characteristics. The dynamic characteristics of the landing impact phase and the skid braking phase are significantly different. If a unified prediction model is used, it is difficult to take into account the prediction accuracy of different phases. (4) In the airborne environment, real-time computing resources are limited. How to meet the real-time requirements while ensuring prediction accuracy is the key issue for realizing airborne deployment.

[0005] Therefore, it is necessary to propose a structural load prediction method based on historical load spectrum test data. By using multi-dimensional feature screening and segmented prediction model construction, combined with airborne edge computing units, the method can realize real-time prediction and output of the three-dimensional structural load of the landing gear, thereby improving the real-time performance and engineering applicability of landing gear structural load monitoring. Summary of the Invention

[0006] The purpose of this invention is to address the problems in existing technologies, such as the difficulty in achieving real-time airborne prediction of landing gear structural loads, the lack of systematic selection of input features, and the difficulty in simultaneously ensuring the accuracy of load prediction at different landing stages. This invention provides an airborne monitoring method and system for landing gear structural loads based on edge computing using a deep learning model. This method constructs a structural load prediction model based on historical load spectrum test data, determines the input feature set of three-dimensional structural loads through multi-dimensional feature filtering, constructs a segmented prediction model by combining the stage-specific features of the landing gear landing process, and deploys the model on an airborne edge computing unit to achieve real-time prediction output of the three-dimensional structural loads of the landing gear.

[0007] To achieve the above objectives, the first aspect of this invention discloses an airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing using a deep learning model, the method comprising: S1. Construct a landing gear structural load prediction model based on historical load spectrum test data; The historical load spectrum test data includes flight parameters, landing gear compression stroke signals, landing gear overload signals, and strain signals at key locations; S2. Perform multi-dimensional feature screening on the historical load spectrum test data; S3. Construct segmented prediction models based on different stages of the landing gear landing process; The segmented prediction model includes a landing impact phase prediction model and a skid braking phase prediction model. S4. Deploy the trained segmented prediction model on the airborne edge computing unit; S5. Collect flight parameters and landing gear monitoring signals within the preset landing event time window; S6. Perform input parameter filtering, data preprocessing, and stage identification on the collected signals, and call the corresponding segmented prediction model based on the identification results; S7. Outputs real-time prediction results of the three-dimensional structural loads of the landing gear.

[0008] According to the method of the first aspect of the present invention, in S2, the multidimensional feature screening specifically includes: Calculate the correlation index between each input parameter and the target load; wherein the correlation index includes at least one of Pearson correlation coefficient, Spearman correlation coefficient or Kendall correlation coefficient; Construct at least one surrogate prediction model and extract importance indicators of input parameters; wherein the surrogate prediction model is a temporal neural network model or a tree-based prediction model, used to evaluate the contribution of each input parameter to the target load prediction result; Normalize the correlation and importance indicators; The normalized indicators are weighted and fused to obtain a comprehensive feature importance index; in the weighting and fusion process, each weight satisfies the non-negativity constraint and the sum of the weights is 1; Based on the comprehensive feature importance index, the input feature parameter sets corresponding to the three-dimensional structural loads are determined by sorting and filtering.

[0009] According to the method of the first aspect of the present invention, in S2, sorting and filtering are performed based on the comprehensive feature importance index, specifically including: Features whose comprehensive feature importance index is higher than a preset threshold are selected; Select the top N features from the sorted list; Select a set of features whose cumulative importance reaches a preset ratio.

[0010] According to the method of the first aspect of the present invention, in S3, the segmented prediction model is constructed based on a time-series deep learning network, the network including a long short-term memory network, a temporal convolutional network, or a combination thereof.

[0011] According to the method of the first aspect of the present invention, in S4, the airborne edge computing unit performs model compression or quantization processing on the prediction model to meet the real-time computing requirements.

[0012] According to the method of the first aspect of the present invention, in S6: The data preprocessing includes outlier removal, filtering, normalization, and time axis alignment. The phase identification distinguishes between the landing impact phase and the skid braking phase by judging the rate of change of compression stroke, peak vertical overload, or change of flight attitude.

[0013] To achieve the above objectives, a second aspect of the present invention discloses an airborne monitoring system for landing gear structural loads of aviation equipment based on edge computing using a deep learning model, the system comprising: The data acquisition module is configured to acquire flight parameters, landing gear compression stroke signals, landing gear overload signals, and strain signals at key locations. The feature selection module is configured to determine the input feature set based on the fusion of correlation analysis and surrogate prediction model importance. The model building module is configured to: build a segmented structure load prediction model based on historical load spectrum data; The edge computing module is configured to perform input parameter filtering, data preprocessing, stage identification, and segmented prediction model invocation. The output module is configured to output real-time prediction results of the three-dimensional structural loads of the landing gear.

[0014] The system according to the second aspect of the present invention: The edge computing module is an embedded low-power processing unit; The output module includes an airborne display interface for displaying load change curves and maximum and minimum load statistics during the landing process; The system also includes a data storage module, configured to store raw monitoring data and load prediction results, and provide a data export interface.

[0015] According to a system of a second aspect of the present invention, the feature selection module specifically includes components configured to: Calculate the correlation index between each input parameter and the target load; wherein the correlation index includes at least one of Pearson correlation coefficient, Spearman correlation coefficient or Kendall correlation coefficient; Construct at least one surrogate prediction model and extract importance indicators of input parameters; wherein the surrogate prediction model is a temporal neural network model or a tree-based prediction model, used to evaluate the contribution of each input parameter to the target load prediction result; Normalize the correlation and importance indicators; The normalized indicators are weighted and fused to obtain a comprehensive feature importance index; in the weighting and fusion process, each weight satisfies the non-negativity constraint and the sum of the weights is 1; Based on the comprehensive feature importance index, the input feature parameter sets corresponding to the three-dimensional structural loads are determined by sorting and filtering.

[0016] According to a system based on a second aspect of the present invention, the edge computing module is configured to perform data preprocessing and stage identification; wherein: The data preprocessing includes outlier removal, filtering, normalization, and time axis alignment. The phase identification distinguishes between the landing impact phase and the skid braking phase by judging the rate of change of compression stroke, peak vertical overload, or change of flight attitude.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) By using a multi-dimensional feature screening mechanism, a comprehensive feature importance index is formed by combining the correlation index and the importance evaluation results of the proxy prediction model, so as to realize the differentiated selection of the input features of the three-dimensional structural load, reduce model redundancy, and improve prediction efficiency. (2) By constructing prediction models for the landing impact stage and the skid-brake stage, segmented predictions for different dynamic stages can be achieved, thereby improving the accuracy of structural load prediction. (3) By deploying the trained segmented prediction model on the airborne edge computing unit, input parameter filtering, data preprocessing and stage identification are completed within the preset landing event time window, so as to realize the real-time prediction output of the three-dimensional structural load of the landing gear. (4) Through modular system structure design, data acquisition, feature filtering, model building, edge computing and output functions are independent of each other and work together, thereby improving the feasibility of system engineering; (5) Under the premise of ensuring prediction accuracy, the system can meet the real-time calculation requirements in the airborne environment by model compression or quantization, thereby enhancing the system’s adaptability to aviation equipment. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This invention illustrates a schematic diagram of the overall architecture for airborne monitoring of landing gear structural loads of aviation equipment based on edge computing using a deep learning model, as shown in an embodiment of the present invention. Figure 2 A flowchart illustrating the airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing of a deep learning model is shown in an embodiment of the present invention. Figure 3 A schematic diagram of the multi-dimensional feature filtering process in an embodiment of the present invention is shown; Figure 4 A schematic diagram of the segmented structure load prediction model in an embodiment of the present invention is shown; Figure 5 A schematic diagram of the airborne edge computing data processing and segmented prediction model invocation process in an embodiment of the present invention is shown; Figure 6 This invention illustrates a schematic diagram of the functional modules of an airborne monitoring system for landing gear structural loads of aviation equipment based on edge computing using a deep learning model, as shown in an embodiment of the present invention. Figure label: 111. Data acquisition module; 112. Feature selection module; 113. Model building module; 114. Edge computing module; 115. Output module; 116. Data storage module. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention discloses an airborne monitoring scheme for landing gear structural loads of aviation equipment based on edge computing using a deep learning model. The overall architecture of the scheme is as follows: Figure 1 As shown, it constructs a structural load prediction model based on historical load spectrum test data, determines the input feature set of three-dimensional structural loads through multi-dimensional feature screening, constructs a segmented prediction model by combining the stage characteristics of the landing gear landing process, and deploys the model on the airborne edge computing unit to realize the real-time prediction output of the three-dimensional structural loads of the landing gear.

[0022] The first aspect of this invention discloses an airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing using a deep learning model.

[0023] (1) A landing gear structural load prediction model is constructed based on historical load spectrum test data. The historical data includes flight parameters, landing gear compression stroke signal, landing gear overload signal and strain signal at key positions.

[0024] (2) Perform multi-dimensional feature filtering on the historical data, including: Calculate the correlation index between each input parameter and the target load; Construct at least one proxy prediction model and extract importance metrics for the input parameters; Normalize the correlation and importance indicators; The normalized indicators are weighted and fused to obtain a comprehensive feature importance index; Based on the comprehensive feature importance index, the input feature parameter sets corresponding to the three-dimensional structural loads are determined by sorting and filtering.

[0025] (3) Construct segmented prediction models based on different stages of the landing gear landing process, including a landing impact stage prediction model and a skid braking stage prediction model.

[0026] (4) Deploy the trained segmented prediction model on the airborne edge computing unit.

[0027] (5) Collect flight parameters and landing gear monitoring signals within the preset landing event time window.

[0028] (6) The collected signals are filtered for input parameters, preprocessed and stage identified, and the real-time prediction results of the three-dimensional structural load of the landing gear are output according to (7).

[0029] Optionally, the correlation index includes at least one of Pearson correlation coefficient, Spearman correlation coefficient, or Kendall correlation coefficient.

[0030] Optionally, the proxy prediction model is a temporal neural network model or a tree-based prediction model, used to evaluate the contribution of each input parameter to the target load prediction result.

[0031] Optionally, in the weighted fusion process, the weights corresponding to each evaluation method satisfy the non-negativity constraint and the sum of the weights is 1.

[0032] Optionally, the input feature filtering is implemented in any of the following ways: Select features whose overall feature importance index is higher than a preset threshold; Select the top N features from the sorted list; Select a set of features whose cumulative importance reaches a preset ratio.

[0033] Optionally, the segmented prediction model is constructed based on a time-series deep learning network, which includes a long short-term memory network, a temporal convolutional network, or a combination thereof.

[0034] Optionally, the phase identification distinguishes between the landing impact phase and the skid braking phase by determining the rate of change of compression stroke, peak vertical overload, or change in flight attitude.

[0035] Optionally, the data preprocessing includes outlier removal, filtering, normalization, and time axis alignment.

[0036] Optionally, the airborne edge computing unit performs model compression or quantization on the prediction model to meet real-time computing requirements.

[0037] The second aspect of this invention discloses an airborne monitoring system for landing gear structural loads of aviation equipment based on edge computing using a deep learning model, comprising: The data acquisition module is used to collect flight parameters, landing gear compression stroke signals, landing gear overload signals, and strain signals at key positions; The feature selection module is used to determine the input feature set based on the fusion of correlation analysis and the importance of the surrogate prediction model; The model building module is used to build a segmented structure load prediction model based on historical load spectrum data; The edge computing module is used to perform input parameter filtering, data preprocessing, stage identification, and segmented prediction model invocation; The output module is used to output the real-time prediction results of the three-dimensional structural loads of the landing gear.

[0038] Optionally, the edge computing module is an embedded low-power processing unit.

[0039] Optionally, the output module includes an airborne display interface for displaying the load change curve and the statistical results of the maximum and minimum loads during the landing process.

[0040] Optionally, the system further includes a data storage module for storing raw monitoring data and load prediction results, and provides a data export interface.

[0041] The first embodiment of the present invention is described below. This embodiment provides an airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing using a deep learning model, such as... Figure 2 As shown, the specific process includes the following steps.

[0042] S100. A landing gear structural load prediction model is constructed based on historical load spectrum test data.

[0043] In this embodiment, the landing gear structure of an aircraft is selected as the monitoring object. Historical load spectrum test data are collected, including flight parameters, landing gear compression stroke signals, landing gear overload signals, and strain signals at key locations. This data is used to construct the training dataset for a three-dimensional structural load prediction model of the landing gear.

[0044] S200. Perform multi-dimensional feature filtering on historical data (specific steps are as follows) Figure 3 (As shown).

[0045] S210. Calculate the correlation index between each input parameter and the target load.

[0046] The correlation indicators include Pearson correlation coefficient, Spearman correlation coefficient, or Kendall correlation coefficient, which are used to measure the degree of correlation between input parameters and triaxial structural loads.

[0047] S220. Construct at least one proxy prediction model and extract importance indicators of the input parameters.

[0048] The proxy prediction model is a temporal neural network model or a tree-based prediction model, used to evaluate the contribution of each input parameter to the target load prediction result. In this embodiment, it is LSTM, TCN and XGBoost.

[0049] S230. Normalize the correlation and importance indicators.

[0050] Evaluation indicators from different sources are numerically normalized to ensure comparability between different evaluation methods.

[0051] S240. Weighted fusion of the normalized indicators yields a comprehensive feature importance index.

[0052] In the weighted fusion process, the weights corresponding to each evaluation method satisfy the non-negativity constraint and the sum of the weights is 1.

[0053] S250. Sort and filter according to the comprehensive feature importance index, and determine the set of input feature parameters corresponding to the three-dimensional structural loads respectively.

[0054] Input feature filtering methods include: 1) Select features whose overall feature importance index is higher than a preset threshold; 2) Select the features of the top N items in the sort; 3) Select a set of features whose cumulative importance reaches a preset ratio.

[0055] In this embodiment, the first method is used, with a threshold of 0.7.

[0056] Through the above steps, the input feature sets corresponding to the three structural loads are determined.

[0057] S300: Construct segmented prediction models based on different stages of the landing gear landing process.

[0058] like Figure 4 As shown, this embodiment constructs a segmented prediction model based on different stages of the landing gear landing process, including: (1) Landing impact phase prediction model; (2) Prediction model for the skid-brake phase.

[0059] The segmented prediction model is built based on a time-series deep learning network, which includes a long short-term memory network, a temporal convolutional network, or a combination thereof.

[0060] S400: Deploy the trained segmented prediction model on the airborne edge computing unit.

[0061] In this embodiment, the trained segmented prediction model is deployed to an airborne edge computing unit. The airborne edge computing unit can perform model compression or quantization on the prediction model to meet real-time computing requirements.

[0062] The S500 collects flight parameters and landing gear monitoring signals within a preset landing event time window.

[0063] During actual flight, flight parameters and landing gear monitoring signals are collected within a preset landing event time window. These monitoring signals include landing gear compression stroke signals, landing gear overload signals, and strain signals at critical positions. This time window covers the landing gear impact phase and the taxiing and braking phase.

[0064] S600 performs input parameter filtering, data preprocessing, and stage identification on the acquired signals, and calls the corresponding segmented prediction model based on the identification results.

[0065] like Figure 5 As shown, the following steps are performed within the airborne edge computing unit.

[0066] Input parameter filtering.

[0067] Based on the triaxial structural load input feature set determined in step S200, the real-time acquired signals are filtered.

[0068] Data preprocessing.

[0069] Data preprocessing includes outlier removal, filtering, normalization, and time axis alignment.

[0070] Phase identification.

[0071] The landing impact phase and the skid braking phase are distinguished by judging the rate of change of compression stroke, peak vertical overload, or changes in flight attitude.

[0072] Call the corresponding segmented prediction model.

[0073] Based on the phase identification results, the landing impact phase prediction model or the skid braking phase prediction model is invoked to calculate the load.

[0074] S700 outputs real-time prediction results of the three-dimensional structural loads of the landing gear.

[0075] The system ultimately outputs real-time predictions of the three-dimensional structural loads on the landing gear. The output module can display the load variation curves and maximum and minimum load statistics during landing via an onboard display interface. The system also includes a data storage module for storing raw monitoring data and load prediction results, and provides a data export interface.

[0076] Through the above steps, real-time airborne prediction of three-dimensional structural loads of landing gear based on edge computing using a deep learning model can be achieved, improving the real-time performance and engineering applicability of landing gear structural load monitoring.

[0077] The second embodiment of the present invention is described below. This embodiment provides an airborne monitoring system for the structural load of aircraft landing gear based on edge computing using a deep learning model. This monitoring system is an airborne monitoring system for the structural load of aircraft landing gear based on edge computing using a deep learning model, such as... Figure 6 As shown, the airborne monitoring system 110 for landing gear structure load of aviation equipment based on edge computing of deep learning model includes a data acquisition module 111, a feature selection module 112, a model building module 113, an edge computing module 114, an output module 115, and a data storage module 116.

[0078] The data acquisition module 111 is used to collect flight parameters, landing gear compression stroke signals, landing gear overload signals, and strain signals at key positions. The data acquisition module collects relevant landing gear monitoring signals within a preset landing event time window, providing input data for subsequent load prediction.

[0079] The feature selection module 112 is used to determine the input feature set based on the fusion of correlation analysis and the importance of the surrogate prediction model. The feature selection module performs a multi-dimensional feature selection process, including calculating the correlation index between the input parameters and the target load, constructing a surrogate prediction model and extracting the importance index, normalizing the correlation and importance indices, and weighting and fusing the normalized indices to obtain a comprehensive feature importance index, thereby determining the input feature parameter set corresponding to the three-dimensional structural loads.

[0080] The model building module 113 is used to establish a segmented structural load prediction model based on historical load spectrum data. The segmented prediction model includes a landing impact stage prediction model and a skid-braking stage prediction model, and the model is built based on a time-series deep learning network.

[0081] The edge computing module 114 is used to perform input parameter filtering, data preprocessing, stage identification, and segmented prediction model invocation. The edge computing module performs input parameter filtering, outlier removal, filtering, normalization, and time axis alignment on the real-time acquired signals. It identifies the landing impact stage and the ski-running braking stage by judging the compression stroke change rate, vertical overload peak value, or flight attitude change, and then invokes the corresponding segmented prediction model to calculate the three-dimensional structural load. The edge computing module is an embedded low-power processing unit and can perform model compression or quantization on the prediction model to meet real-time computing requirements.

[0082] Output module 115 is used to output real-time prediction results of the three-dimensional structural loads of the landing gear. The output module includes an airborne display interface for displaying the load change curves and statistical results of the maximum and minimum loads during the landing process.

[0083] The data storage module 116 is used to store raw monitoring data and load prediction results, and provides a data export interface for subsequent data analysis and management.

[0084] This embodiment achieves real-time airborne prediction and output of three-dimensional structural loads on the landing gear through the collaborative work of the data acquisition module, feature filtering module, model building module, edge computing module, and output module. While ensuring prediction accuracy, multi-dimensional feature filtering reduces model input redundancy, segmented prediction models improve load prediction accuracy at different stages, and the edge computing module enables real-time calculation and result display, thereby enhancing the system's engineering applicability and real-time response capabilities.

[0085] Compared with the prior art, the present invention has the following beneficial effects: (1) By using a multi-dimensional feature screening mechanism, a comprehensive feature importance index is formed by combining the correlation index and the importance evaluation results of the proxy prediction model, so as to realize the differentiated selection of the input features of the three-dimensional structural load, reduce model redundancy, and improve prediction efficiency. (2) By constructing prediction models for the landing impact stage and the skid-brake stage, segmented predictions for different dynamic stages can be achieved, thereby improving the accuracy of structural load prediction. (3) By deploying the trained segmented prediction model on the airborne edge computing unit, input parameter filtering, data preprocessing and stage identification are completed within the preset landing event time window, so as to realize the real-time prediction output of the three-dimensional structural load of the landing gear. (4) Through modular system structure design, data acquisition, feature filtering, model building, edge computing and output functions are independent of each other and work together, thereby improving the feasibility of system engineering; (5) Under the premise of ensuring prediction accuracy, the system can meet the real-time calculation requirements in the airborne environment by model compression or quantization, thereby enhancing the system’s adaptability to aviation equipment.

[0086] Please note that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification. The above embodiments only illustrate several implementations of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be pointed out that for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the appended claims.

Claims

1. An airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing using a deep learning model, characterized in that, The method includes: S1. Construct a landing gear structural load prediction model based on historical load spectrum test data; The historical load spectrum test data includes flight parameters, landing gear compression stroke signals, landing gear overload signals, and strain signals at key locations; S2. Perform multi-dimensional feature screening on the historical load spectrum test data; S3. Construct segmented prediction models based on different stages of the landing gear landing process; The segmented prediction model includes a landing impact phase prediction model and a skid braking phase prediction model. S4. Deploy the trained segmented prediction model on the airborne edge computing unit; S5. Collect flight parameters and landing gear monitoring signals within the preset landing event time window; S6. Perform input parameter filtering, data preprocessing, and stage identification on the collected signals, and call the corresponding segmented prediction model based on the identification results; S7. Outputs real-time prediction results of the three-dimensional structural loads of the landing gear.

2. The airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing of a deep learning model according to claim 1, characterized in that, In S2, multidimensional feature selection specifically includes: Calculate the correlation index between each input parameter and the target load; wherein the correlation index includes at least one of Pearson correlation coefficient, Spearman correlation coefficient or Kendall correlation coefficient; Construct at least one surrogate prediction model and extract importance indicators of input parameters; wherein the surrogate prediction model is a temporal neural network model or a tree-based prediction model, used to evaluate the contribution of each input parameter to the target load prediction result; Normalize the correlation and importance indicators; The normalized indicators are weighted and fused to obtain a comprehensive feature importance index; in the weighting and fusion process, each weight satisfies the non-negativity constraint and the sum of the weights is 1; Based on the comprehensive feature importance index, the input feature parameter sets corresponding to the three-dimensional structural loads are determined by sorting and filtering.

3. The airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing of a deep learning model according to claim 2, characterized in that, In S2, sorting and filtering are performed based on the comprehensive feature importance index, specifically including: Features whose comprehensive feature importance index is higher than a preset threshold are selected; Select the top N features from the sorted list; Select a set of features whose cumulative importance reaches a preset ratio.

4. The airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing of a deep learning model according to claim 1, characterized in that, In S3, the segmented prediction model is built based on a time-series deep learning network, which includes a long short-term memory network, a temporal convolutional network, or a combination thereof.

5. The airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing of a deep learning model according to claim 1, characterized in that, In S4, the airborne edge computing unit performs model compression or quantization on the prediction model to meet real-time computing requirements.

6. The airborne monitoring method for landing gear structural loads of aviation equipment based on edge computing of a deep learning model according to claim 1, characterized in that, In S6: The data preprocessing includes outlier removal, filtering, normalization, and time axis alignment. The phase identification distinguishes between the landing impact phase and the skid braking phase by judging the rate of change of compression stroke, peak vertical overload, or change of flight attitude.

7. An airborne monitoring system for landing gear structural loads of aviation equipment based on edge computing using a deep learning model, characterized in that, The system includes: The data acquisition module is configured to acquire flight parameters, landing gear compression stroke signals, landing gear overload signals, and strain signals at key locations. The feature selection module is configured to determine the input feature set based on the fusion of correlation analysis and surrogate prediction model importance. The model building module is configured to: build a segmented structure load prediction model based on historical load spectrum data; The edge computing module is configured to perform input parameter filtering, data preprocessing, stage identification, and segmented prediction model invocation. The output module is configured to output real-time prediction results of the three-dimensional structural loads of the landing gear.

8. The airborne monitoring system for landing gear structural loads of aviation equipment based on edge computing of a deep learning model according to claim 7, characterized in that: The edge computing module is an embedded low-power processing unit; The output module includes an airborne display interface for displaying load change curves and maximum and minimum load statistics during the landing process; The system also includes a data storage module, configured to store raw monitoring data and load prediction results, and provide a data export interface.

9. The airborne monitoring system for landing gear structural loads of aviation equipment based on edge computing of a deep learning model according to claim 7, characterized in that, The feature filtering module is specifically configured as follows: Calculate the correlation index between each input parameter and the target load; wherein the correlation index includes at least one of Pearson correlation coefficient, Spearman correlation coefficient or Kendall correlation coefficient; Construct at least one surrogate prediction model and extract importance indicators of input parameters; wherein the surrogate prediction model is a temporal neural network model or a tree-based prediction model, used to evaluate the contribution of each input parameter to the target load prediction result; Normalize the correlation and importance indicators; The normalized indicators are weighted and fused to obtain a comprehensive feature importance index; in the weighting and fusion process, each weight satisfies the non-negativity constraint and the sum of the weights is 1; Based on the comprehensive feature importance index, the input feature parameter sets corresponding to the three-dimensional structural loads are determined by sorting and filtering.

10. An airborne monitoring system for landing gear structural loads of aviation equipment based on edge computing of a deep learning model according to claim 7, characterized in that, The edge computing module is configured to perform data preprocessing and stage identification; wherein: The data preprocessing includes outlier removal, filtering, normalization, and time axis alignment. The phase identification distinguishes between the landing impact phase and the skid braking phase by judging the rate of change of compression stroke, peak vertical overload, or change of flight attitude.

Images