Aircraft brake system evaluation and prediction method based on fault simulation and geometric characteristics

Abstract

A method for evaluating and predicting aircraft brake systems based on fault simulation and geometric features belongs to the field of electrical data processing technology. The method includes: constructing a simulation model of the aircraft brake system, injecting faults into the simulation model according to different levels of fault injection control parameters to generate a degradation monitoring dataset for the system's entire life cycle; dividing the degradation monitoring dataset into a training dataset and a test dataset; using the training dataset to train a GoogLeNet model to obtain a health assessment model; using the test dataset to assess the health of the aircraft brake system; constructing a topological graph based on multiple aircraft brake system operation data and health; clustering the nodes in the topological graph into K clusters; establishing a remaining life prediction model for each cluster; performing similarity matching between samples and the K clusters, and selecting the best-matching remaining life prediction model to predict the remaining life of the system. The present invention can accurately predict the remaining life of an aircraft brake system.

Description

Technical Field

[0001] The invention relates to an aircraft brake system evaluation and prediction method based on fault simulation and geometric characteristics, and belongs to the technical field of electrical data processing. Background Art

[0002] With the advancement of modern aviation technology, aircraft brake systems, as a crucial component for ensuring the safe completion of ground-phase missions, are crucial for flight safety. Timely and accurate health assessment and remaining life prediction of aircraft brake systems not only effectively prevents accidents and ensures flight safety, but also provides strong technical support for aircraft maintenance and support planning. Current technologies for brake system health assessment and remaining life prediction primarily include expert knowledge-based approaches and data-driven approaches. Expert knowledge-based approaches primarily leverage the experience and rules of domain experts, combined with selected brake system performance monitoring parameters, to determine the health status of the system. However, these approaches are limited by their heavy reliance on expert knowledge, the difficulty in constructing a rule-based knowledge base, and their lack of flexibility and generalizability. Another category of approaches involves data-driven health assessment and remaining life prediction. Given the difficulty in collecting brake system data and obtaining full-life degradation data, generative model-based approaches mitigate data scarcity by generating samples similar to actual degradation data. However, these approaches place high demands on the quality of the raw data, requiring sufficient monitoring data from different degradation stages to generate more comprehensive and accurate data. Summary of the Invention

[0003] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide an aircraft brake system evaluation and prediction method based on fault simulation and geometric features. The purpose of the invention is to solve the problem of inaccurate remaining life prediction of aircraft brake systems due to the difficulty in collecting brake system data and obtaining full life degradation data.

[0004] To overcome the shortcomings of the prior art, the present invention aims to provide an aircraft brake system evaluation and prediction method based on fault simulation and geometric features, which comprises:

[0005] Step 1: Build a simulation model of the aircraft brake system, obtain a set of typical failure modes, select fault injection control parameters based on the failure modes in the typical failure mode set, and inject faults into the simulation model according to different degrees of the selected fault injection control parameters to generate a degradation monitoring dataset for the entire life cycle of the aircraft brake system.

[0006] Step 2: Divide the degradation monitoring dataset into a training dataset and a test dataset;

[0007] Step 3: Obtain a training fusion feature set from the training dataset and use it to train the GoogLeNet model to obtain a health assessment model. Obtain a test geometry feature set from the test dataset and use the test geometry feature set to evaluate the health of the aircraft brake system.

[0008] Step 4: Construct a topology graph based on the full-life geometric characteristics of N aircraft brake system operation data and health assessment results; cluster the nodes in the topology graph into K clusters, where K is greater than or equal to 3; establish a remaining life prediction model for each cluster; obtain sample graph node features from the test sample, perform similarity matching between the sample graph node features and the K clusters, and select the remaining life prediction model established by the best matching cluster to predict the remaining life of the aircraft brake system, where N is a positive integer greater than or equal to 2.

[0009] Compared with the existing technology, the aircraft brake system evaluation and prediction method based on fault simulation and geometric features provided by the present invention solves the problem of inaccurate aircraft brake system health status monitoring caused by the difficulty in collecting brake system data and obtaining full life degradation data through the above technical solution, thereby accurately predicting the remaining life of the aircraft brake system. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a flow chart of the aircraft brake system evaluation and prediction method based on fault simulation and geometric features provided by the present invention.

[0011] Figure 2 This is a graph showing the relationship between the sensor drift fault injection control current and the braking time provided by the present invention.

[0012] Figure 3 This is a graph showing the relationship between the injected brake pressure and the braking time in the case of a sensor drift fault provided by the present invention.

[0013] Figure 4 This is a graph showing the relationship between the sensor drift fault injection wheel speed and the braking time provided by the present invention.

[0014] Figure 5 This is a relationship diagram of the fault injection control current of the electro-hydraulic servo valve provided by the present invention versus the braking time.

[0015] Figure 6 This is a graph showing the relationship between the injection brake pressure and the braking time when a fault occurs in the electro-hydraulic servo valve provided by the present invention.

[0016] Figure 7 This is a graph showing the relationship between the speed of the wheel injected into the electro-hydraulic servo valve and the braking time provided by the present invention. DETAILED DESCRIPTION

[0017] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0018] In the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance.

[0019] Figure 1 This is a flow chart of the aircraft brake system evaluation and prediction method based on fault simulation and geometric features provided by the present invention, such as Figure 1 As shown, the present invention provides an aircraft brake system evaluation and prediction method based on fault simulation and geometric features, which includes:

[0020] Step 1: Build a simulation model of the aircraft brake system, obtain a set of typical failure modes, select fault injection control parameters based on the failure modes in the typical failure mode set, and inject faults into the simulation model according to different degrees of the selected fault injection control parameters to generate a degradation monitoring dataset for the entire life cycle of the aircraft brake system.

[0021] Step 2: Divide the degradation monitoring dataset into a training dataset and a test dataset;

[0022] Step 3: Obtain a training fusion feature set from the training dataset and use it to train the GoogLeNet model to obtain a health assessment model. Obtain a test geometry feature set from the test dataset and use the test geometry feature set to evaluate the health of the aircraft brake system.

[0023] Step 4: Construct a topology graph based on the full-life geometric characteristics of N aircraft brake system operation data and health assessment results; cluster the nodes in the topology graph into K clusters, where K is greater than or equal to 3; establish a remaining life prediction model for each cluster; obtain sample graph node features from the test sample, perform similarity matching between the sample graph node features and the K clusters, and select the remaining life prediction model established by the best matching cluster to predict the remaining life of the aircraft brake system, where N is a positive integer greater than or equal to 2.

[0024] In step 1, the simulation model of the aircraft brake system includes the simulation model of the aircraft brake system, the wheel motion model, the hydraulic system model and the brake device model. Among them, the body mechanics model is used to analyze the forces in the heading direction and vertical direction of the aircraft; the wheel motion model is mainly used to analyze the bonding force at the contact point between the tire and the ground, and the corresponding bonding torque; the hydraulic system model is used to analyze the automatic control characteristics of the current feedback loop based on the speed signal; and the brake device model is used to analyze the braking capacity.

[0025] In step 1, the selection of typical fault modes within the typical fault mode set requires combining the actual use of the aircraft brake system with some expert knowledge to select faults with a high probability of occurrence and that can be simulated by the simulation model. Examples of these fault modes include sensor drift and electro-hydraulic servo valve failure. The core principle of sensor drift failure lies in a discrepancy between the wheel speed data transmitted by the wheel speed sensor and the actual wheel speed data. This discrepancy increases with the severity of the fault. Electro-hydraulic servo valve failure manifests itself primarily as a weakening of the valve's ability to control the hydraulic system due to aging, leading to system failure. For these two fault modes, control current, brake pressure, wheel speed, and braking time are selected as fault characterization parameters. These four parameters can, to a certain extent, reflect the overall health of the aircraft brake system, thus providing strong support for evaluation and prediction.

[0026] In step 1, the selection of fault injection control parameters requires consideration of the operating mechanism of typical fault modes and the selection of appropriate control parameter variables. Ultimately, based on the selected fault injection control parameters, varying their values ​​can be used to achieve varying degrees of fault injection, thereby simulating the lifecycle degradation of the aircraft brake system and obtaining lifecycle degradation monitoring data. For example, the greater the deviation between the wheel speed sensor and the actual wheel speed data, the more severe the fault, while the weaker the electro-hydraulic servo valve's control over the hydraulic system, the more severe the fault. Therefore, increasing the gain coefficient in the control system increases the severity of sensor drift faults, while decreasing the gain coefficient increases the severity of electro-hydraulic servo valve faults. Therefore, using the gain coefficient as a fault injection control parameter, different gain coefficients are selected to simulate faults in the corresponding wheel speed sensor and electro-hydraulic servo valve for the two fault modes, and the injected gain coefficient is used as a quantitative representation of the fault severity.

[0027] In step 1, the fault injection control parameter is recorded as C. Each time the C value is set, a fault simulation is completed. Assume that there is a set of fault injection control parameters This means that the aircraft fails after running N sorties, that is, a total of N simulations are performed, and the performance degradation data set of the entire life cycle is finally obtained. , where the data set obtained by the nth simulation is , N is a positive integer greater than or equal to 5, and M is a positive integer greater than or equal to 2.

[0028] For example, Figure 2 、 Figure 3 and Figure 4As shown in the figure, for sensor drift faults, the sensor drift fault gain coefficient range is 1-1.34. A gain coefficient of 1.0 produces the healthiest data for the aircraft's brake system, indicating a brake system health of 1.0. Similarly, a sensor drift fault gain coefficient of 1.34 results in a brake system health of 0. Gain coefficients between 1.18 and 1.34 represent the fault range.

[0029] like Figure 5 、 Figure 6 and Figure 7 As shown in the figure, for an electro-hydraulic servo valve fault, a sensor drift fault gain coefficient range of 1-0.7 is selected. A gain coefficient of 1.0 yields the healthiest aircraft brake system data, indicating a brake system health of 1.0. Similarly, a sensor drift fault gain coefficient of 0.7 results in a brake system health of 0. The fault range is between 0.7 and 0.82.

[0030] Since the failure characteristics of the two failure modes are different, in order to obtain more comprehensive degradation data throughout the entire life cycle, it was finally determined that for sensor drift failure, the total number of simulations was 341 times, and the number of simulation data was 341 sets; for electro-hydraulic servo valve failure, the total number of simulations was 251, and the number of simulation data obtained was 251 sets.

[0031] In step 2, the training set ratio r is selected to divide the performance degradation data set D of the entire life cycle into the training data set and test dataset , , For example, for the health assessment task, r=0.7, that is, the data of 238 operating flights are shuffled as the training set, and the remaining 113 sets of operating flight data are used as the test set; for the remaining life prediction task, r=0.6, that is, the data of 150 operating flights are used as the training set, and the remaining 101 sets of data are used as the test set.

[0032] In step 2, the training data set and test dataset The data in the training data set are normalized for subsequent processing. The normalization process is, for example, maximum-minimum normalization. For example, The normalization of the data in includes:

[0033] Get the training dataset The maximum value of the data and minimum value , then the normalized training data set , in .

[0034] In step 3, the fusion features in the training fusion feature set are the concatenation of the training geometric features extracted from the degradation monitoring data in the training dataset and the high-dimensional robust health representation extracted from the training geometric features.

[0035] Taking the extraction of geometric features from monitoring data in the training dataset as an example, the extraction of geometric features includes:

[0036] S3-1: The time domain information of the monitoring data trajectory formed by the degradation monitoring data in the degradation monitoring data set of the aircraft brake system is aggregated by a sliding window to obtain a time domain indicator sequence. Specifically, the normalized training data set Perform principal component extraction to obtain the principal component normalized training data set , cut with a sliding window of width W and step size ST For each segment, extract the time domain index to obtain the time domain index sequence;

[0037] S3-2: Split the time domain indicator sequence into P segments using the inflection point as the segmentation point, perform trend modeling on the time domain indicator sequence of each segment to obtain a time domain indicator curve for each segment, where P is a positive integer greater than or equal to 2;

[0038] S3-3: Extract the training geometric features of the aircraft brake system based on the time domain indicator curve of each segment, and analyze the changing patterns of the training geometric features of the aircraft brake system to fit the health degradation characteristics of the aircraft brake system.

[0039] The geometric features include, for example, the average slope of the curve, the area of ​​the curve, etc. The calculation formula of the average slope of the curve is as shown in Formula 1:

[0040] Formula 1,

[0041] Where, represents the extracted value of the fault characterization parameter at the end time T of the pth segment, represents the extracted value of the fault characterization parameter at the start time 0 of the pth segment, and T represents the total duration of the segment.

[0042] The curve area calculation formula is as shown in Formula 2:

[0043] Formula 2,

[0044] Where, represents the extracted value of the fault characterization parameter at time t.

[0045] Finally, the geometric features of the pth segment are: .

[0046] High-dimensional robust health representations are extracted from geometric features using a high-dimensional robust health representation extraction model. The high-dimensional robust health representation extraction model is trained using a compression denoising autoencoder model. The training process includes:

[0047] S3-4: Obtain the coding characteristics of the aircraft braking system according to Formula 3:

[0048] , Formula 3,

[0049] Where, is the activation function of the encoder; and is the parameter of the encoder; is the geometric characteristics of the pth segment of the aircraft brake system Add noise features;

[0050] S3-5: Obtain the decoder reconstruction features according to formula 4:

[0051] Formula 4,

[0052] Where, is the activation function of the decoder; and Parameters for the decoder;

[0053] S3-6: Based on reconstruction features and geometric features Calculate the first loss function value ;

[0054] S3-7: Determine the value of the first loss function Is it the smallest? If so, the feature output by the encoder is a high-dimensional robust health representation; if not, modify the parameters 、 、 and value, and then returns to step S3-4.

[0055] Finally, the fusion features of the p-th fragment are obtained: ,

[0056] Where, Represents splicing, the fusion features of P fragments are spliced ​​together to obtain the training fusion features .

[0057] In step 3, the health assessment model includes a feature extraction module and a degradation state measurement module; the feature extraction module is configured to obtain performance degradation features based on the fusion features; the degradation state measurement module is configured to obtain the health of the aircraft brake system based on the performance degradation features.

[0058] The feature extraction module is configured to obtain the performance degradation feature according to the fusion feature, including:

[0059] S3-8: Perform initial convolution and maximum pooling on the fusion feature using formula 5 to obtain the first feature :

[0060] Formula 5,

[0061] Where, Represents the training fusion feature; represents the convolution weight, Represents the convolution bias term; represents convolution; represents maximum pooling;

[0062] S3-9: For the first feature The second feature set is obtained by performing multi-scale feature extraction through formula 6 , I is a positive integer greater than or equal to 2; for the first feature The third feature set is obtained by multi-scale pooling through formula 7 , J is a positive integer greater than or equal to 1; preferably, I=4, J=1, wherein,

[0063] Formula 6,

[0064] Where, The convolution kernel is Convolution; Indicates activation; The convolution kernel is Convolution;

[0065] Formula 7,

[0066] Where, Represents the core Max pooling;

[0067] S3-10: Concatenate all the second features in the second feature set to obtain the first feature vector, concatenate the third features in the third feature set to obtain the second feature vector, and then concatenate the first feature vector and the second feature vector to obtain the performance degradation feature. .

[0068] The degradation state measurement module is configured to obtain the health of the aircraft braking system according to the performance degradation characteristics, including:

[0069] S3-11: The fourth characteristic is obtained according to the performance degradation characteristic using formula 8:

[0070] Formula 8,

[0071] Where, Indicates degenerate features; represents the first fully connected weight, represents the first fully connected bias term; is the activation function;

[0072] S3-12: Based on the fourth characteristic, the predicted health of the aircraft brake system for the nth flight is obtained using Formula 9: Formula 9,

[0073] Where, represents the second fully connected weight, represents the second fully connected bias term; Represents a mapping function. The value range is [0,1]. The closer the healthiness is to 1, the healthier the system is, and the closer it is to 0, the more severely degraded the system is.

[0074] Optionally, obtaining a predicted health H of the aircraft brake system through steps 3-12;

[0075] S3-13: Label the health of the brake system of the nth aircraft according to the fault injection control parameters as follows:

[0076] Formula 10,

[0077] In the above formula is the fault injection control parameter for the nth flight, The fault injection control parameter sequence for all fault simulation injection flights, is the true health label of the nth flight:

[0078] S3-14: Construct the second loss function according to formula 11:

[0079] Formula 11,

[0080] In the above formula, is the number of training samples (the number of training flights);

[0081] S3-15: Determine the second loss function Is it the smallest? If so, output the best parameters. 、 、 、 Otherwise, adjust the parameters 、 、 、 And return to step S3-11.

[0082] The present invention trains the GoogLeNet model into a health assessment model through steps S3-4 to S2-15.

[0083] In step 3, the health status of the brake system is measured by the health assessment model, which specifically includes the geometric characteristics of the test sample By inputting the health assessment model, the health status assessment results of each test sample can be obtained. The present invention uses the mean absolute error to measure the accuracy of the model, which is calculated as shown in Formula 12:

[0084] Formula 12,

[0085] In the above formula, , are the predicted value and true value of the health of the brake system obtained based on test sample i, respectively.

[0086] Step 4 aims to use the geometric feature data of the aircraft brake system's operating data and the health assessment results to construct a topological graph structure, which will be used in the subsequent remaining life prediction model to mine the life change pattern of the aircraft brake system from the graph data, thereby achieving a more accurate remaining life prediction.

[0087] Step 4 includes:

[0088] S4-1: Construct a topological graph based on the full-life geometric feature data and health assessment results of N flight operations of the aircraft brake system. This graph is then used by the subsequent remaining life prediction model to mine the life variation patterns of the aircraft brake system from the graph data, thereby achieving more accurate remaining life prediction.

[0089] S4-2: Based on the constructed health status topology graph, the N flight samples are clustered using a graph structure clustering method. An independent RUL prediction model is constructed within each cluster, thereby achieving a more refined RUL prediction model.

[0090] S4-3: Extract sample graph node features from the sample data, perform similarity matching on these features with K clusters, select the RLS model of the most similar cluster, and use it to predict the RLS of the aircraft brake system. This method enables large-sample migration modeling based on health state similarity, improving prediction accuracy and robustness.

[0091] Step S4-1 specifically includes:

[0092] S4-1-1: Consider each flight as a node in the graph. The characteristics of each node include geometric characteristics and brake system health. The above two types of features are spliced ​​together to form the graph node features: ,

[0093] Where: is the graph node feature; represents geometric features; H represents health assessment results; Represents a vector concatenation operation.

[0094] Similarly, geometric features are extracted for the test samples, and the health of the brake system is evaluated based on the test samples. The geometric features and health evaluation results are spliced ​​together to obtain the sample graph node features. ,The graph node features comprehensively reflect the original degradation trend and deep ,potential feature structure of the brake system, which is beneficial to ,improving the accuracy of subsequent remaining life prediction.

[0095] S4-1-2: Cluster the nodes in the topology graph into K clusters, and connect the nodes in the K clusters through undirected edges to obtain a health status topology graph G=(V,E), where V represents the set of all flight nodes, E represents the edge set constructed based on feature similarity, and K is greater than or equal to 3; establish a remaining life prediction model for each cluster.

[0096] Step S4-2 specifically includes:

[0097] S4-2-1: Divide the nodes into K clusters Each cluster Representing a set of flight data with similar structure and health status trends in the health status space, providing a basis for the construction and training of subsequent remaining life prediction models;

[0098] S4-2-2: Build a remaining life prediction model, which is trained by a convolutional attention network (Attention-CNN);

[0099] S4-2-3: Use the remaining life prediction model to predict the remaining life of the aircraft brake system.

[0100] In the present invention, the training process of training the convolutional attention network into a remaining life prediction model includes:

[0101] S4-2-2-1: Input the graph node features into the input layer of the convolutional neural network and extract the hidden features from the output layer , according to the implicit features Compute the query vector Q, key vector K, and value vector V:

[0102] ; ; ;

[0103] Where, 、 and are the weights of the query vector Q, key vector K and value vector respectively; 、 and They are the query vector Q, key vector K and bias of value vector respectively;

[0104] S4-2-2-2: Get the attention features based on the query vector Q, key vector K and value vector V: ;

[0105] Where T represents transpose, is the training gradient, p is the number of training times;

[0106] S4-2-2-3: Input the attention feature into the fully connected layer to build a mapping relationship with the remaining lifespan. The calculation process is as follows:

[0107] ,

[0108] Where, is the predicted value of the remaining life of the aircraft brake system for the nth flight, and are the weight vector and bias vector of the fully connected layer respectively;

[0109] S4-2-2-4: Constructing the remaining lifetime label based on the fault injection control parameters:

[0110] ,

[0111] In the above formula is the fault injection control parameter for the nth flight, Fault injection control parameter sequence for all fault simulation injection flights, The actual remaining service life label of the aircraft brake system for the nth flight;

[0112] S4-2-2-5: Construct the third loss function according to the following formula:

[0113] ,

[0114] In the above formula, is the number of training samples (the number of training flights);

[0115] S-2-2-6: Determine the third loss function Is it the smallest? If so, output the best parameters. 、 、 、 、 、 、 、 , otherwise adjust the parameters by the reverse gradient method and return to step S4-2-2-1.

[0116] The present invention trains the convolutional attention network into a remaining life prediction model through steps S4-2-2-1 to S4-2-2-6.

[0117] In the present invention, the remaining life prediction model is used to predict the remaining life of the brake system, specifically including:

[0118] S4-2-3-1: For the test sample to be predicted, obtain its graph node features , computational graph node features The similarity between each cluster center, let cluster The set of feature vectors of each cluster center is: , then the cluster closest to the test sample is calculated as follows:

[0119] ,

[0120] Where dist(·) represents the Euclidean distance metric function;

[0121] S4-2-3-2: Determined cluster matching clusters After that, the graph node features Input to Cluster The input layer of the remaining life prediction model of the aircraft brake system is output as the output layer. , calculate the fourth loss function according to formula 12:

[0122] Formula 12,

[0123] In the above formula, , are the actual value and predicted value of the remaining service life of the aircraft brake system obtained based on test sample i.

[0124] Finally, the above steps can complete the health assessment and remaining life prediction of all test samples.

[0125] The present invention also provides an aircraft brake system evaluation and prediction system based on fault simulation and geometric features, which includes a storage medium and one or more processors. The storage medium stores a computer program, and the computer program is called by the one or more processors to implement the above-mentioned aircraft brake system evaluation and prediction method based on fault simulation and geometric features.

[0126] The present invention also provides a computer program product, which utilizes a computer language to compile the above-mentioned aircraft brake system evaluation and prediction method based on fault simulation and geometric features into a computer program that is called and executed by one or more processors.

[0127] In the present invention, one or a combination of multiple steps mentioned above still falls within the scope of the present invention.

[0128] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. An aircraft brake system evaluation and prediction method based on fault simulation and geometric features, characterized in that: include: Step 1: Build a simulation model of the aircraft brake system, obtain a set of typical failure modes, select fault injection control parameters based on the failure modes in the typical failure mode set, and inject faults into the simulation model according to different degrees of the selected fault injection control parameters to generate a degradation monitoring dataset for the entire life cycle of the aircraft brake system. Step 2: Divide the degradation monitoring dataset into a training dataset and a test dataset; Step 3: Obtain a training fusion feature set from the training dataset and use it to train the GoogLeNet model to obtain a health assessment model. Obtain a test geometry feature set from the test dataset and use the test geometry feature set to evaluate the health of the aircraft brake system. The fusion features in the training fusion feature set are the concatenation of the training geometric features extracted from the degradation monitoring data in the training dataset and the high-dimensional robust health representation extracted from the training geometric features; Step 4: Construct a topology graph based on the full-life geometric characteristics of N aircraft brake system operation data and health assessment results; cluster the nodes in the topology graph into K clusters, where K is greater than or equal to 3; establish a remaining life prediction model for each cluster; obtain sample graph node features from the test sample, perform similarity matching between the sample graph node features and the K clusters, and select the remaining life prediction model established by the best matching cluster to predict the remaining life of the aircraft brake system, where N is a positive integer greater than or equal to 2.

2. The aircraft brake system evaluation and prediction method based on fault simulation and geometric features according to claim 1 is characterized in that: The extraction of training geometric features includes: S3-1: performing time domain information aggregation on the monitoring data trajectory formed by the degradation monitoring data in the degradation monitoring data set of the aircraft brake system through a sliding window to obtain a time domain indicator sequence; S3-2: Segment the time domain indicator sequence using the inflection point as the segmentation point to obtain P segments, perform trend modeling on the time domain indicator sequence of each segment to obtain a time domain indicator curve for each segment, where P is a positive integer greater than or equal to 2; S3-3: Extract the training geometric features of the aircraft brake system based on the time domain indicator curve of each segment, and analyze the change pattern of the training geometric features of the aircraft brake system to fit the health state degradation characteristics of the aircraft brake system.

3. The aircraft brake system evaluation and prediction method based on fault simulation and geometric features according to claim 2 is characterized in that: The high-dimensional robust health representation is extracted from the training geometric features of the aircraft brake system using a high-dimensional robust health representation extraction model. The high-dimensional robust health representation extraction model is trained using a compression denoising autoencoder model. The training process includes: S3-4: Obtain the coding characteristics of the aircraft brake system according to the following formula: , , where is the activation function of the encoder; and is the parameter of the encoder; is the geometric characteristics of the pth segment of the aircraft brake system Add noise features; S3-5: Obtain the decoder reconstruction features according to the following formula: , Where, is the activation function of the decoder; and Parameters for the decoder; S3-6: Based on reconstruction features and geometric features Calculate the first loss function value ; S3-7: Determine the value of the first loss function Is it the smallest? If so, the feature output by the encoder is a high-dimensional robust health representation; if not, modify the parameters 、 、 and value, and then returns to step S3-4.

4. The aircraft brake system evaluation and prediction method based on fault simulation and geometric features according to claim 1 is characterized in that: The health assessment model includes a feature extraction module and a degradation state measurement module; the feature extraction module is configured to obtain performance degradation features based on training fusion features; The degradation state measurement module is configured to obtain the health of the aircraft braking system according to the performance degradation characteristics.

5. The aircraft brake system evaluation and prediction method based on fault simulation and geometric features according to claim 4 is characterized in that: The feature extraction module is configured to obtain performance degradation features based on the training fusion features, including: S3-8: Perform initial convolution and maximum pooling on the training fusion features to obtain the first feature : , Where, Represents the training fusion feature; represents the convolution weight, Represents the convolution bias term; represents convolution; represents maximum pooling; S3-9: For the first feature Perform multi-scale feature extraction to obtain the second feature set , I is a positive integer greater than or equal to 2; for the first feature Perform multi-scale pooling to obtain the third feature set , J is a positive integer greater than or equal to 1; S3-10: splicing the second features in the second feature set to obtain a first splicing vector, splicing the third features in the third feature set to obtain a second splicing vector, and then splicing the first splicing vector and the second splicing vector to obtain a performance degradation feature.

6. The aircraft brake system evaluation and prediction method based on fault simulation and geometric features according to claim 5 is characterized in that: ， Where, The convolution kernel is Convolution; Indicates activation; The convolution kernel is Convolution; , where Represents the core The maximum pooling of .

7. The aircraft brake system evaluation and prediction method based on fault simulation and geometric features according to claim 6 is characterized in that: The degradation state measurement module is configured to obtain the health of the aircraft braking system according to the performance degradation characteristics, including: S3-11: Obtain the fourth feature based on the performance degradation feature: , Where, Indicates performance degradation characteristics; represents the first fully connected weight, represents the first fully connected bias term, is the activation function; S3-12: Obtain the predicted health of the aircraft brake system for the nth flight based on the fourth feature: Where, represents the second fully connected weight, represents the second fully connected bias term; Represents a mapping function.

8. The aircraft brake system evaluation and prediction method based on fault simulation and geometric features according to claim 1 is characterized in that: The remaining lifespan prediction model is trained by a convolutional attention network.

9. The aircraft brake system evaluation and prediction method based on fault simulation and geometric features according to any one of claims 1 to 8, characterized in that: The simulation model of the aircraft braking system includes an airframe mechanics model, a wheel motion model, a hydraulic system model, and a brake device model. The airframe mechanics model is used to analyze the forces in the aircraft's heading and vertical directions; the wheel motion model is used to analyze the bonding force at the tire-ground contact point, as well as the corresponding bonding torque; the hydraulic system model is used to analyze the automatic control characteristics of the current feedback loop based on the speed signal; and the brake device model is used to analyze the braking capability.

Images