Aircraft dangerous working condition screening method and system

Abstract

The invention belongs to the technical field of aircraft dangerous working condition screening, and provides an aircraft dangerous working condition screening method and system, and the method comprises the steps: carrying out the analysis of extracted data through a plurality of machine learning and statistical algorithms, determining the danger degree of each force element, and carrying out the sorting, and obtaining a sorting result of a plurality of working conditions; performing coarse verification and fine verification on the sorting result of each working condition to realize screening; a plurality of clustering algorithms are utilized to classify all working conditions to obtain different clustering results, and the accuracy of the clustering results is evaluated through cross validation to obtain a proper clustering algorithm and a proper clustering result; and fusing the clustering results, and matching the screened working condition sorting result with the clustering results to obtain a working condition screening result. Through fusion of various algorithms and activation functions, the application of the verification layer can automatically adjust the use condition of the algorithms, and the adaptability and the accuracy are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of aircraft dangerous operating condition screening, and in particular relates to an aircraft dangerous operating condition screening method and system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] The load envelope method or design point method is usually used to screen dangerous aircraft conditions, and there is subjectivity, which may lead to inconsistency and deviation in the evaluation results. Many existing methods may encounter difficulties in calculation and analysis when dealing with large and complex systems. Some methods are only applicable to simple, linear relationships and have difficulty in dealing with complex, dynamic flight environments. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a method and system for screening dangerous operating conditions of aircraft. The present invention performs random sampling and post-clustering after feature extraction to enhance its generalization ability. It can automatically adjust the algorithm usage through the fusion of multiple algorithms and the application of activation functions and verification layers, thereby improving the generalization ability, adaptability and accuracy.

[0005] According to some embodiments, the present invention adopts the following technical solutions: A method for screening dangerous operating conditions of an aircraft comprises the following steps: Obtain the six force elements, resultant force, resultant moment and stress value of rough model condition of the cross section; Extract the various forces and moments from the acquired data, visualize them, and calculate relevant mathematical indicators of the extracted data; Combined with the above-mentioned relevant mathematical indicators, the extracted data is analyzed using a variety of machine learning and statistical algorithms, the danger level of each working condition is evaluated from different perspectives, and the conditions are ranked according to the danger level scores; Perform a rough verification on the sorting results of each working condition, compare the sorting results of various working conditions with the sorting order of the stress values ​​of the rough model working conditions, and if the comparison difference of the sorting results of the working conditions of an algorithm exceeds the set value, delete the sorting results of the working conditions corresponding to the algorithm; The sorting results of each working condition are verified in detail to determine the accuracy of the sorting results of each working condition after deletion, which is used as an indicator for activating the function switch. The sorting results of the working conditions corresponding to the algorithms with accuracy less than the set value are deleted; By weighted soft voting, the remaining working condition ranking results are ranked according to the accuracy to obtain a final working condition ranking result; According to the idea of ​​random forest ensemble, the characteristics of each working condition are randomly sampled; Using a variety of clustering algorithms, the features of random sampling of all working conditions are classified to obtain different clustering results. The accuracy of the clustering results is evaluated through cross-validation to obtain the appropriate clustering algorithm and clustering results. The appropriate clustering algorithm is integrated with the clustering results to determine the final classification results. According to the number of working conditions in each classification, the final working condition ranking results are divided from safe to dangerous, corresponding to the clustering results; Dangerous working conditions are screened out based on the final classification results and the final working condition ranking results after division.

[0006] As an optional implementation, the relevant mathematical indicators include maximum and minimum values, evaluation values, standard deviation, skewness and kurtosis.

[0007] As an optional implementation, in the process of analyzing the extracted data using a variety of machine learning and statistical algorithms, principal component analysis (PCA), singular value decomposition (SVD), topsis, grey correlation analysis, standard correlation analysis (CRITIC), rank sum ratio method, standard correlation analysis-additive regression model method (CRITIC-GARP), standard correlation analysis-topsis and standard correlation analysis-entropy weight method (CRITIC-Entropy) are used for analysis.

[0008] As a further limitation, the analysis processes of the CRITIC-GARP method, the CRITIC-Topsis method and the CRITIC-Entropy method are respectively: The CRITIC method is used to calculate the weight of each indicator. CRITIC considers the correlation and contrast between indicators and assigns weights. An additive regression model is constructed to describe the impact of each indicator on the degree of risk. The predicted value of the additive regression model is used to generate a comprehensive score, and finally the ranking is achieved. The CRITIC method is used to calculate the weight of each indicator, analyze the correlation and contrast between indicators, standardize the data of different dimensions, determine the ideal solution and negative ideal solution based on the standardized data, use the Topsis algorithm to calculate the distance from each evaluation object to the ideal solution and negative ideal solution, calculate the comprehensive score based on the distance and weight, and finally achieve ranking; The CRITIC method is used to calculate weights based on correlation and contrast, and the entropy weight method is used to calculate the information entropy of each indicator to consider the information content of the indicator. The weights obtained by the CRITIC method are combined with the results of the entropy weight method to obtain the final weights. The weights are used to perform weighted synthesis of the data of each indicator to achieve final ranking.

[0009] As an optional implementation, the rough model working condition stress values ​​are sorted by searching for the number of units greater than a threshold value in the target part of the aircraft and performing the sorting.

[0010] As an optional implementation method, the process of rough verification is to use straight line fitting for comparison, with the stress rank as the y-axis and the ranks of the working condition ranking results obtained by different algorithms as the x-axis, to perform scatter point fitting, and determine the difference between the slope of the straight line and the preset best fitting value. If the difference exceeds the set threshold, the working condition ranking results obtained by the algorithm are deleted.

[0011] As an optional implementation method, the detailed verification process is to perform a position difference comparison, differencing the rank of the working condition ranking results obtained by the two algorithms for each working condition, setting the working conditions whose position difference exceeds a predetermined proportion of the total number of working conditions to 1, and determining the ratio of the number of working conditions set to 1 to the total number of working conditions. If the ratio exceeds the set value, the working condition ranking results obtained by the algorithm are deleted.

[0012] As an optional implementation, the clustering algorithm includes a neural network-based clustering algorithm, a density-based clustering algorithm, a hierarchical clustering algorithm and a partitioning clustering algorithm.

[0013] As an optional implementation, the process of evaluating the accuracy of clustering results through cross-validation includes: obtaining the macro recall rate and macro F1 coefficient through cross-validation to evaluate the accuracy of clustering; and then blocking inappropriate algorithms through activation functions to finally obtain a suitable clustering algorithm and clustering results.

[0014] As an optional implementation, the process of fusing the appropriate clustering algorithm with the clustering result includes normalizing the recall rate and the F1 coefficient, adding them, and normalizing them again to obtain the voting weight of each clustering algorithm, and fusing them by stacking.

[0015] As an optional implementation, the classification includes four types, namely safe, relatively safe, relatively dangerous and dangerous.

[0016] An aircraft dangerous operating condition screening system, comprising: A data acquisition module is configured to acquire six force elements and resultant forces, resultant moments, and stress values ​​of a rough model working condition of a cross section; an input module configured to extract various forces and moments from the acquired data, visualize them, and calculate relevant mathematical indicators of the extracted data; a sorting module configured to analyze the extracted data using a variety of machine learning and statistical algorithms in combination with the relevant mathematical indicators, evaluate the degree of danger of each working condition from different perspectives, sort the conditions according to the degree of danger score, and sort the conditions to obtain sorting results for a variety of working conditions; A rough verification module is configured to perform rough verification on the sorting results of each working condition, compare the sorting results of various working conditions with the sorting order of the stress values ​​of the rough model working conditions, and if the comparison difference of the sorting results of the working conditions of an algorithm exceeds a set value, the sorting results of the working conditions corresponding to the algorithm are deleted; A detailed verification module is configured to perform detailed verification on the sorting results of each working condition, determine the accuracy of the sorting results of each working condition after deletion, and use it as an indicator of the activation function switch to delete the sorting results of the working condition corresponding to the algorithm whose accuracy is less than the set value; A weighted sorting module is configured to sort the remaining working condition sorting results according to the accuracy by weighted soft voting to obtain a final working condition sorting result; The data augmentation module is configured to randomly sample the features of each working condition based on the idea of ​​random forest ensemble; The clustering module is configured to classify the features randomly sampled from all working conditions using a variety of clustering algorithms to obtain different clustering results, evaluate the accuracy of the clustering results through cross-validation, and obtain a suitable clustering algorithm and clustering results; The integration module is configured to fuse the appropriate clustering algorithm with the clustering result to determine the final classification result, and to classify the final working condition ranking result from safe to dangerous according to the number of working conditions in each classification, corresponding to the clustering result; The output module is configured to screen out dangerous working conditions according to the final classification result and the final working condition sorting result after division.

[0017] Compared with the prior art, the present invention has the following beneficial effects: In the preliminary sorting process, the present invention can completely extract the characteristics of each working condition through sorting by multiple algorithms; since some sorting in each method is different, algorithm integration is performed to obtain a high-accuracy sorting; different algorithms are adapted to problems in different environments, and the combination of multiple algorithms can enhance the generalization ability of the integration.

[0018] The present invention can be used on a variety of machine models. Secondly, a variety of different algorithms are adopted, and verification layers and activation functions are set, so that it can be applicable to problems in a variety of environments; through the fusion of multiple algorithms and the application of activation functions and verification layers, the algorithm usage can be automatically adjusted.

[0019] According to the idea of ​​random forest integration, the present invention randomly samples the characteristics of each working condition and constructs different sample data sets for different clustering algorithms to increase the differences between clusters, improve the accuracy of algorithm integration and the generalization ability of the model. The present invention utilizes clustering algorithms of different categories to further increase the differences among base learners, provides multiple clustering algorithm options for different data sets, enhances generalization capabilities, obtains multiple clustering results, and fuses them through stacking to improve the accuracy and reliability of clustering results. The present invention has high accuracy and can combine multiple weaker learning bases into a strong learning base; the present invention has strong interpretability and is different from the black box effect of neural networks.

[0020] The present invention effectively screens load conditions, quickly identifies relatively safe and dangerous conditions, and effectively reduces the time for identifying and optimizing dangerous loads; Compared with the previous critical load identification process, this algorithm automates data analysis and has the advantages of integrated learning, parallel execution of algorithms, and time saving.

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0023] Figure 1 is a schematic diagram comparing the results of ranking the dangerousness of working conditions using different algorithms in an embodiment; Figure 2 The present invention is a flow chart of a method for screening dangerous operating conditions of an aircraft in an embodiment. DETAILED DESCRIPTION

[0024] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0025] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0026] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0027] In the absence of conflict, the embodiments in this application and the features in the embodiments may be combined with each other.

[0028] Embodiment 1 A method for screening dangerous aircraft operating conditions, such as Figure 2 As shown, the following steps are included: Obtain the six force elements, resultant force, resultant moment and stress value of rough model condition of the cross section; Extract the various forces and moments from the acquired data, visualize them, and calculate relevant mathematical indicators of the extracted data; Combined with the above-mentioned relevant mathematical indicators, the extracted data is analyzed using a variety of machine learning and statistical algorithms, the danger level of each working condition is evaluated from different perspectives, and the conditions are ranked according to the danger level scores; Perform a rough verification on the sorting results of each working condition, compare the sorting results of various working conditions with the sorting order of the stress values ​​of the rough model working conditions, and if the comparison difference of the sorting results of the working conditions of an algorithm exceeds the set value, delete the sorting results of the working conditions corresponding to the algorithm; The sorting results of each working condition are verified in detail to determine the accuracy of the sorting results of each working condition after deletion, which is used as an indicator for activating the function switch. The sorting results of the working conditions corresponding to the algorithms with accuracy less than the set value are deleted; By weighted soft voting, the remaining working condition ranking results are ranked according to the accuracy to obtain a final working condition ranking result; According to the idea of ​​random forest ensemble, the characteristics of each working condition are randomly sampled; Using a variety of clustering algorithms, the features of random sampling of all working conditions are classified to obtain different clustering results. The accuracy of the clustering results is evaluated through cross-validation to obtain the appropriate clustering algorithm and clustering results. The appropriate clustering algorithm is integrated with the clustering results to determine the final classification results. According to the number of working conditions in each classification, the final working condition ranking results are divided from safe to dangerous, corresponding to the clustering results; Dangerous working conditions are screened out based on the final classification results and the final working condition ranking results after division.

[0029] In this embodiment, the relevant mathematical indicators include maximum and minimum values, evaluation values, standard deviation, skewness and kurtosis.

[0030] In this embodiment, in the process of analyzing the extracted data using a variety of machine learning and statistical algorithms, PCA principal component analysis method, SVD singular value decomposition method, Topsis method, grey correlation analysis method, CRITIC analysis method, rank sum ratio method, CRITIC-GARP method, CRITIC-Topsis method and CRITIC-Entropy method are used for analysis.

[0031] The analysis processes of the CRITIC-GARP method, CRITIC-Topsis method and CRITIC-Entropy method are respectively: The CRITIC method is used to calculate the weight of each indicator. CRITIC considers the correlation and contrast between indicators and assigns weights. An additive regression model is constructed to describe the impact of each indicator on the degree of risk. The predicted value of the additive regression model is used to generate a comprehensive score, and finally the ranking is achieved. The CRITIC method is used to calculate the weight of each indicator, analyze the correlation and contrast between indicators, standardize the data of different dimensions, determine the ideal solution and negative ideal solution based on the standardized data, use the Topsis algorithm to calculate the distance from each evaluation object to the ideal solution and negative ideal solution, calculate the comprehensive score based on the distance and weight, and finally achieve ranking; The CRITIC method is used to calculate weights based on correlation and contrast, and the entropy weight method is used to calculate the information entropy of each indicator to consider the information content of the indicator. The weights obtained by the CRITIC method are combined with the results of the entropy weight method to obtain the final weights. The weights are used to perform weighted synthesis of the data of each indicator to achieve final ranking.

[0032] In this embodiment, the rough model working condition stress value sorting is performed by searching for the number of units greater than a threshold value in the target part of the aircraft and performing the sorting.

[0033] In this embodiment, the process of rough verification is to use straight line fitting for comparison, with the stress rank as the y-axis and the ranks of the working condition ranking results obtained by different algorithms as the x-axis, to perform scatter point fitting, and determine the difference between the slope of the straight line and the preset best fitting value. If the difference exceeds the set threshold, the working condition ranking result obtained by the algorithm is deleted.

[0034] In this embodiment, the detailed verification process is to compare the position differences, and to make a difference in the positions of the working condition ranking results obtained by the two algorithms for each working condition. The working conditions whose position differences exceed a predetermined proportion of the total number of working conditions are set to 1, and the ratio of the number of working conditions set to 1 to the total number of working conditions is determined. If the ratio exceeds the set value, the working condition ranking results obtained by the algorithm are deleted.

[0035] In this embodiment, the clustering algorithm includes a neural network-based clustering algorithm, a density-based clustering algorithm, a hierarchical clustering algorithm, and a partitioning clustering algorithm.

[0036] In this embodiment, the process of evaluating the accuracy of clustering results through cross-validation includes: obtaining the macro recall rate and macro F1 coefficient through cross-validation to evaluate the accuracy of clustering; and then blocking inappropriate algorithms through activation functions to finally obtain a suitable clustering algorithm and clustering results.

[0037] In this embodiment, the process of fusing the appropriate clustering algorithm with the clustering result includes normalizing the recall rate and the F1 coefficient, adding them, and normalizing them again to obtain the voting weight of each clustering algorithm, and fusing them in a stacking manner.

[0038] In this embodiment, the classification includes four types, namely safe, relatively safe, relatively dangerous and dangerous.

[0039] Embodiment 2 An aircraft dangerous operating condition screening system, comprising: A data acquisition module is configured to acquire six force elements and resultant forces, resultant moments, and stress values ​​of a rough model working condition of a cross section; an input module configured to extract various forces and moments from the acquired data, visualize them, and calculate relevant mathematical indicators of the extracted data; a sorting module configured to analyze the extracted data using a variety of machine learning and statistical algorithms in combination with the relevant mathematical indicators, evaluate the degree of danger of each working condition from different perspectives, sort the conditions according to the degree of danger score, and sort the conditions to obtain sorting results for a variety of working conditions; A rough verification module is configured to perform rough verification on the sorting results of each working condition, compare the sorting results of various working conditions with the sorting order of the stress values ​​of the rough model working conditions, and if the comparison difference of the sorting results of the working conditions of an algorithm exceeds a set value, the sorting results of the working conditions corresponding to the algorithm are deleted; A detailed verification module is configured to perform detailed verification on the sorting results of each working condition, determine the accuracy of the sorting results of each working condition after deletion, and use it as an indicator of the activation function switch to delete the sorting results of the working condition corresponding to the algorithm whose accuracy is less than the set value; A weighted sorting module is configured to sort the remaining working condition sorting results according to the accuracy by weighted soft voting to obtain a final working condition sorting result; The data augmentation module is configured to randomly sample the features of each working condition based on the idea of ​​random forest ensemble; The clustering module is configured to classify the features randomly sampled from all working conditions using a variety of clustering algorithms to obtain different clustering results, evaluate the accuracy of the clustering results through cross-validation, and obtain a suitable clustering algorithm and clustering results; The integration module is configured to fuse the appropriate clustering algorithm with the clustering result to determine the final classification result, and to classify the final working condition ranking result from safe to dangerous according to the number of working conditions in each classification, corresponding to the clustering result; The output module is configured to screen out dangerous working conditions according to the final classification result and the final working condition sorting result after division.

[0040] Embodiment 3 First, some necessary explanations of terms: Coarse model: In the preliminary design stage of the aircraft, in order to have a preliminary understanding of the aircraft's force response, a coarse model is used for calculation and analysis. Load screening is done on the coarse model. It takes a long time to calculate the force of the detailed model, and load screening cannot be performed based on the detailed model.

[0041] The present embodiment provides a method for screening dangerous operating conditions of an aircraft, which is implemented through a neural network structure. The neural network structure includes an input layer, a sorting layer, a first coarse verification layer, a second fine verification layer, an activation function, a sorting result, a clustering layer, a verification activation layer, a result integration layer, a verification layer, and an output layer, which can realize clustering and screening of dangerous operating conditions.

[0042] The following is a detailed description: (1) Input layer: Get the original data: the six forces and resultant force and moment of the cross section, the stress value of the rough model condition, and sort the magnitude of the rough model stress value. Consider the force and moment as positive indicators, the larger the value, the more dangerous the result.

[0043] In the aircraft model, the aerodynamic load is distributed on each unit node. In order to better characterize the stress condition of the wing, the cross-sectional internal force is used to describe the structural stress.

[0044] The extracted data is visualized through scatter plots and three-dimensional plots. The maximum and minimum values, evaluation values, standard deviations, skewness, and kurtosis of each indicator are calculated to grasp the overall characteristics of the data.

[0045] From now on, the input layer data is ready.

[0046] (2) Sorting layer: The important force and torque data sets in the above correlation analysis are analyzed through machine learning and statistical methods, and objective solution evaluation is made.

[0047] In this embodiment, the method can be selected from the following methods: PCA principal component analysis, SVD singular value decomposition, Topsis method, grey relational degree, CRITIC, rank sum ratio RSR, CRITIC-GARP, CRITIC-Topsis, CRITIC-Entropy.

[0048] In this embodiment, taking the wing as an example, all the force elements and resultant moments of seven sections are analyzed. The seven sections are seven sections evenly distributed from the wing root to the wing tip. The following method is selected for analysis: PCA principal component analysis: center the original data so that the mean of each variable is 0; calculate the covariance matrix of the standardized data and analyze the correlation between variables; solve the eigenvalues ​​and eigenvectors of the covariance matrix. The eigenvalues ​​represent the variance, and the eigenvectors represent the direction of the principal component. Select the first k eigenvalues ​​with large and corresponding eigenvectors to form a new principal component space. Map the original data to the principal component space to obtain the principal component score. Analyze the degree of danger of each factor and rank them according to the principal component score.

[0049] SVD: Prepare the matrix data to be processed (such as the Nexo score matrix), select the top k important features by screening the larger singular values, and reconstruct the matrix with the selected features. This can represent the main information of the original matrix, reduce the dimension, analyze the reconstructed data, and sort the degree of danger based on the corresponding features.

[0050] CRITIC: By considering the correlation and contrast between indicators, the weight of each indicator is calculated for more accurate ranking.

[0051] TOPSIS method: Standardize the evaluation index of each force element. Use the given weights to weight the standardized data. The ideal solution is the optimal value of each indicator, and the negative ideal solution is the worst value of each indicator; calculate the Euclidean distance of each force element to the ideal solution and the negative ideal solution respectively. Calculate the proximity of each force element to the ideal solution through the distance. Sort by proximity value to get the degree of danger.

[0052] Grey correlation: Establish comparison sequence and reference sequence. Normalize the data for comparison; determine the correlation between each force element and the target by calculating the grey correlation between the standardized sequences; set the grey correlation threshold to filter out the force elements with high correlation; sort the force elements by the degree of danger according to the correlation value; calculate the order and overall ratio of different force elements, and then sort them.

[0053] CRITIC-GARP: First, the CRITIC method is used to calculate the weight of each indicator. CRITIC takes into account the correlation and contrast between indicators, so that the weights can be allocated more reasonably. An additive regression model is constructed to describe the impact of each indicator on the target (degree of danger). The predicted value of the model is used to generate a comprehensive score, and finally the ranking is achieved.

[0054] CRITIC-TOPSIS: The CRITIC method is used to calculate the weight of each indicator, and the method analyzes the correlation and contrast between indicators. The data of different dimensions are standardized to facilitate the calculation of TOPSIS. According to the standardized data, the ideal solution and the negative ideal solution are determined. The distance from each evaluation object to the ideal solution and the negative ideal solution is calculated. The comprehensive score is calculated based on the distance and weight, and finally the ranking is achieved.

[0055] CRITIC-Entropy: First, the CRITIC method is used to calculate the weights based on correlation and contrast. Second, the entropy weight method is used to calculate the information entropy of each indicator to consider the information content of the indicator. The weights obtained by CRITIC are combined with the results of the entropy weight method to obtain the final weights. The weights are used to perform a weighted synthesis of the data of each indicator and finally achieve ranking.

[0056] This embodiment can completely extract the characteristics of each working condition by sorting multiple algorithms; since some sortings in each method are different, algorithm integration is performed to obtain a high-accuracy sorting; different algorithms are suitable for problems in different environments, and the combination of multiple algorithms can enhance the generalization ability of the integration.

[0057] So far, we have obtained a variety of sorting methods formed by multiple algorithms.

[0058] (3) The first coarse verification layer: The ranking of dangerous working conditions obtained by various algorithms is compared with the ranking of stress values ​​of rough model working conditions, such as Figure 1 The sorting of the stress values ​​of the rough model condition is as follows: searching for the number of units greater than the threshold value in a certain part and sorting them.

[0059] Verification method: straight line fitting comparison, the slope of the straight line fitting in the best similarity case is 1; The stress ranking can be used as the y-axis, and the ranking of the working condition ranking results obtained by different algorithms can be used as the x-axis for scatter point fitting. After scatter point fitting, the difference between the slope of the straight line and 1 is observed. If the results of a certain algorithm are greatly different, it will be screened out.

[0060] The process of sorting the stress values ​​of the rough model condition includes: (a) Obtain the stress values ​​of all elements of a certain structure in the aircraft finite element model; (b) Setting multiple stress value thresholds and calculating the number of elements greater than these thresholds; (c) setting the importance coefficient for each threshold; (d) The number of cells greater than a certain threshold is multiplied by the corresponding threshold importance coefficient to obtain the risk score; (e) Add up the danger level scores of each threshold to obtain the final danger level score of the working condition.

[0061] Compare the two sorting results and perform screening.

[0062] (4) Second fine validation layer (two-by-two cross validation): Determine the accuracy of the algorithm under this dataset as an indicator for switching the activation function. Filter out algorithms that are completely unsuitable for this structure / problem environment.

[0063] Verification method: compare the position difference. Difference is made between the two algorithms for each working condition. The working condition where the position difference exceeds 5% of the total number of working conditions is set to 1.

[0064] The similarity is the number of working conditions set to 1 / the total number of working conditions. If the similarity exceeds the set value, the working condition sorting result obtained by the algorithm is deleted.

[0065] (5) Activation function: sigmoid activation function. Through the activation function, algorithms with high accuracy can be activated and passed downward, blocking algorithms with accuracy lower than the threshold. This activation function is continuous and differentiable, which is conducive to back propagation. By screening out algorithms with low accuracy, the sorting results entering the next layer are determined. At this point, the remaining sorting results enter the next round.

[0066] (6) Sorting results: According to the soft voting method in ensemble learning, the accuracy of each sorting algorithm after the activation function is different, that is, the weight of its influence on the result is different. The final result of the dangerousness ranking is obtained through weighted soft voting.

[0067] (7) Clustering layer: This includes partitioning clustering, hierarchical clustering, density-based clustering, neural network-based clustering, and multiple dimensional methods. Different types of clustering algorithms further increase the differences between base learners, provide multiple clustering algorithm options for different data sets, and enhance generalization capabilities. Multiple clustering results are obtained.

[0068] In this embodiment, all working conditions are classified into four categories: safe, relatively safe, relatively dangerous, and dangerous.

[0069] Partitioning Clustering: Partitioning clustering works by dividing the data into a fixed number of clusters, usually using K-means. Its goal is to minimize the differences between data points within each cluster while maximizing the differences between different clusters.

[0070] Hierarchical clustering: Organize data by building a hierarchical tree structure, which can be divided into bottom-up aggregation and top-down splitting methods.

[0071] Density-based clustering: It identifies clusters by analyzing the density of points and can effectively handle clusters of arbitrary shapes and noise. This algorithm does not require the number of clusters to be specified in advance and is suitable for complex data sets.

[0072] Neural network-based clustering: Use deep learning models to extract low-dimensional feature representations of the data, followed by traditional clustering.

[0073] (8) Validation and activation layer: Different clustering algorithms are adapted to different environments. The macro recall rate and macro F1 coefficient are obtained through cross-validation, which can roughly evaluate the accuracy of clustering. The activation function is then used to block inappropriate algorithms, and finally a suitable clustering algorithm and clustering results are obtained.

[0074] Macro recall: In a multi-class classification problem, the recall of each class is calculated and then the recall of all classes is averaged.

[0075] Macro F1 coefficient: It is the harmonic mean of recall and precision.

[0076] (9) Result integration layer: The stacking method is used to perform fusion to improve the accuracy and reliability of clustering results. The recall rate and F1 coefficient are normalized, added, and normalized again to obtain the voting weight of each clustering algorithm. The difference between the stacking method and the algorithm fusion is that the clustering category probability is determined by the algorithm recall rate and F1 coefficient.

[0077] Stacking: By calculating the recall rate and F1 coefficient of each algorithm as above, these indicators are normalized to obtain their voting weights. Then, according to the output label and weight of each algorithm, weighted voting is performed on the data points, and finally a label is determined for each data point. Finally, the clustering results after fusion are evaluated to see whether the performance has been improved.

[0078] (10) Verification layer: Based on the final results, four categories are obtained, with different numbers in each category. Based on the number of working conditions in each category and the final result, the danger level is ranked from safe to dangerous, corresponding to the clustering results.

[0079] After clustering integration, the clustering results (several categories) are obtained. At this time, it is not known which category is more dangerous. Combined with the sorting results of all working conditions, these categories are arranged from safe to dangerous to filter out dangerous working conditions.

[0080] (11) Output layer: After multiple layers of verification, the four classification results and the ranking of the degree of danger are finally output, and the dangerous working conditions are screened out to complete the task. The screening of dangerous working conditions of the aircraft is realized.

[0081] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of one or more computer-usable storage media (including but not limited to disk storage, CD - ROM , optical storage, etc.).

[0082] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0083] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0084] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present invention without creative labor shall be included in the protection scope of the present invention.

Claims

1. A method for screening dangerous operating conditions of an aircraft, characterized in that: The following steps are involved: Obtain the six force elements, resultant force, resultant moment and stress value of rough model condition of the cross section; Extract the various forces and moments from the acquired data, visualize them, and calculate the relevant mathematical indicators of the extracted data; Combined with the above-mentioned relevant mathematical indicators, the extracted data is analyzed using a variety of machine learning and statistical algorithms, the danger level of each working condition is evaluated from different perspectives, and the conditions are ranked according to the danger level scores; Perform a rough verification on the sorting results of each working condition, compare the sorting results of various working conditions with the sorting order of the stress values ​​of the rough model working conditions, and if the comparison difference of the sorting results of the working conditions of an algorithm exceeds the set value, delete the sorting results of the working conditions corresponding to the algorithm; The sorting results of each working condition are verified in detail to determine the accuracy of the sorting results of each working condition after deletion, which is used as an indicator for activating the function switch. The sorting results of the working conditions corresponding to the algorithms with accuracy less than the set value are deleted; By weighted soft voting, the remaining working condition ranking results are ranked according to the accuracy to obtain a final working condition ranking result; According to the idea of ​​random forest ensemble, the characteristics of each working condition are randomly sampled; Using a variety of clustering algorithms, the features of random sampling of all working conditions are classified to obtain different clustering results. The accuracy of the clustering results is evaluated through cross-validation to obtain the appropriate clustering algorithm and clustering results. The appropriate clustering algorithm is integrated with the clustering results to determine the final classification results. According to the number of working conditions in each classification, the final working condition ranking results are divided from safe to dangerous, corresponding to the clustering results; Dangerous working conditions are screened out based on the final classification results and the final working condition ranking results after division.

2. A method for screening dangerous operating conditions of an aircraft as claimed in claim 1, characterized in that: The relevant mathematical indicators include maximum and minimum values, evaluation values, standard deviation, skewness and kurtosis.

3. A method for screening dangerous operating conditions of an aircraft as claimed in claim 1, characterized in that: In the process of analyzing the extracted data using a variety of machine learning and statistical algorithms, principal component analysis, singular value decomposition, approximate ideal solution ranking method, grey correlation analysis, standard correlation analysis, rank sum ratio method, standard correlation analysis-additive regression model method, standard correlation analysis-approximate ideal solution ranking method and standard correlation analysis-entropy weight method were used for analysis.

4. A method for screening dangerous operating conditions of an aircraft as claimed in claim 3, characterized in that: The analysis processes of the standard correlation analysis-additive regression model method, the standard correlation analysis-approximate ideal solution sorting method and the standard correlation analysis-entropy weight method are respectively: The weight of each indicator is calculated using the standard correlation analysis method, which considers the correlation and contrast between indicators and assigns weights; an additive regression model is constructed to describe the impact of each indicator on the degree of risk, and the predicted value of the additive regression model is used to generate a comprehensive score, and finally achieve ranking; The standard correlation analysis method is used to calculate the weight of each indicator, analyze the correlation and contrast between indicators, standardize the data of different dimensions, determine the ideal solution and negative ideal solution based on the standardized data, and use the approximate ideal solution sorting method to calculate the distance of each evaluation object to the ideal solution and negative ideal solution. The comprehensive score is calculated based on the distance and weight, and finally the sorting is achieved; The standard correlation analysis method is used to calculate the weights based on correlation and contrast, and the entropy weight method is used to calculate the information entropy of each indicator to consider the information content of the indicator. The weights obtained by the standard correlation analysis method are combined with the results of the entropy weight method to obtain the final weights. The weights are used to perform weighted synthesis of the data of each indicator to achieve final ranking.

5. The method for screening dangerous operating conditions of an aircraft as claimed in claim 1, characterized in that: The rough model working condition stress value sorting is performed by searching for the number of units greater than a threshold value at the target part of the aircraft and performing the sorting.

6. A method for screening dangerous operating conditions of an aircraft as claimed in claim 1, characterized in that: The process of rough verification is to use straight line fitting for comparison, with stress ranking as the y-axis and the ranking of the working condition ranking results obtained by different algorithms as the x-axis, to perform scatter point fitting, and determine the gap between the slope of the straight line and the preset best fitting value. If the gap exceeds the set threshold, the working condition ranking results obtained by the algorithm are deleted.

7. A method for screening dangerous operating conditions of an aircraft as claimed in claim 1, characterized in that: The detailed verification process is to compare the position differences, subtract the positions of the working condition ranking results obtained by the two algorithms for each working condition, set the working conditions whose position differences exceed a predetermined proportion of the total number of working conditions to 1, and determine the ratio of the number of working conditions set to 1 to the total number of working conditions. If the ratio exceeds the set value, the working condition ranking results obtained by the algorithm are deleted.

8. The method for screening dangerous operating conditions of an aircraft as claimed in claim 1, characterized in that: The clustering algorithms include a neural network-based clustering algorithm, a density-based clustering algorithm, a hierarchical clustering algorithm, and a partitioning clustering algorithm; The process of evaluating the accuracy of clustering results through cross-validation includes: obtaining macro recall and macro F1 coefficient through cross-validation to evaluate the accuracy of clustering; Then, the activation function is used to block inappropriate algorithms, and finally the appropriate clustering algorithm and clustering results are obtained.

9. The method for screening dangerous operating conditions of an aircraft as claimed in claim 1, characterized in that: The process of fusing the appropriate clustering algorithm with the clustering results includes normalizing the recall rate and the F1 coefficient, adding them, and normalizing them again to obtain the voting weight of each clustering algorithm, and fusing them using the stacking generalization method.

10. An aircraft dangerous operating condition screening system, characterized in that: include: A data acquisition module is configured to acquire six force elements and resultant forces, resultant moments, and stress values ​​of a rough model working condition of a cross section; an input module configured to extract various forces and moments from the acquired data, visualize them, and calculate relevant mathematical indicators of the extracted data; a sorting module configured to analyze the extracted data using a variety of machine learning and statistical algorithms in combination with the relevant mathematical indicators, evaluate the degree of danger of each working condition from different perspectives, sort the conditions according to the degree of danger score, and sort the conditions to obtain sorting results for a variety of working conditions; A rough verification module is configured to perform rough verification on the sorting results of each working condition, compare the sorting results of various working conditions with the sorting order of the stress values ​​of the rough model working conditions, and if the comparison difference of the sorting results of the working conditions of an algorithm exceeds a set value, the sorting results of the working conditions corresponding to the algorithm are deleted; A detailed verification module is configured to perform detailed verification on the sorting results of each working condition, determine the accuracy of the sorting results of each working condition after deletion, and use it as an indicator of the activation function switch to delete the sorting results of the working condition corresponding to the algorithm whose accuracy is less than the set value; A weighted sorting module is configured to sort the remaining working condition sorting results according to the accuracy by weighted soft voting to obtain a final working condition sorting result; The data augmentation module is configured to randomly sample the features of each working condition based on the idea of ​​random forest ensemble; The clustering module is configured to classify the features randomly sampled from all working conditions using a variety of clustering algorithms to obtain different clustering results, evaluate the accuracy of the clustering results through cross-validation, and obtain a suitable clustering algorithm and clustering results; The integration module is configured to fuse the appropriate clustering algorithm with the clustering result to determine the final classification result, and to classify the final working condition ranking result from safe to dangerous according to the number of working conditions in each classification, corresponding to the clustering result; The output module is configured to screen out dangerous working conditions according to the final classification result and the final working condition sorting result after division.

Images