A method for generating aircraft flight simulation parameter data

Abstract

The application provides a kind of generation method of aircraft flight simulation parameter data, comprising: obtaining a plurality of formats of aircraft flight parameter data source file, and preprocessing according to different formats;Construct a generation network and a discrimination network;Multiple random distribution vectors are input into the generation network, the generation network receives the feedback of the discrimination network, and the generation network parameters are updated;The transition virtual data output by the generation network is sent into the discrimination network, the discrimination network receives the feedback of the generation network and the preprocessed data source file, outputs the image authenticity degree label, and updates the discrimination network parameters;Multiple training samples are used for training;The trained generation model is used to test the generation network, output virtual aircraft flight parameter data, and finally converted into a plurality of formats of virtual flight parameter data file.The application can generate a large number of reliable data meeting the use requirements through relatively fewer aircraft flight parameter data source files of multiple formats.

Description

Technical Field

[0001] This invention relates to the field of airborne maintenance systems for civil aircraft, and specifically to a method for generating aircraft flight simulation parameter data. Background Technology

[0002] Fault diagnosis in civil aircraft airborne maintenance systems involves judging potential or already occurring faults based on sensor data, condition monitoring information, known structural characteristics and parameters, environmental conditions, and operational history. This includes determining the fault's gender, category, severity, cause, and location, and indicating its occurrence and development trend. Both knowledge-based and data-driven fault diagnosis methods require a foundation of long-term, large-scale fault information data. In fault prediction, data-driven prediction methods need to learn the relationship between input and output from a large amount of historical data, and reliability life prediction methods based on improved Weibull distributions require the collection of large amounts of reliability data, which is characterized by its authenticity, randomness, and comprehensiveness. In health assessment, statistical models require a large amount of monitoring data, historical data, and condition data. Similarly, aircraft maintenance research based on big data, cloud computing, data mining, and deep learning also requires a large amount of diverse and characteristic data as a research foundation.

[0003] In reality, collecting a large amount of reliable and valid data that meets the required characteristics is time-consuming and labor-intensive, requiring a huge amount of work to collect, and the collection process is difficult to be accurate, complete, timely, effective and economical. Summary of the Invention

[0004] In view of this, the present application provides a method for generating aircraft flight simulation parameter data, which can generate a large amount of reliable data that meets certain requirements, thus solving the problem of the huge amount of work required to collect data.

[0005] This application provides the following technical solution: a method for generating aircraft flight simulation parameter data, comprising:

[0006] Acquire aircraft flight parameter data source files in various formats;

[0007] The acquired data source files are preprocessed according to different formats. The preprocessing includes classifying the data files, analyzing the characteristics of the file data, and transforming the data into data graphs and images in various ways.

[0008] Constructing a generative network and a discriminative network;

[0009] Multiple randomly distributed vectors are input into the generator network, which simultaneously receives feedback from the discriminator network, thereby updating the generator network parameters.

[0010] The transitional aircraft flight parameter data output by the generator network is fed into the discriminator network. The discriminator network simultaneously receives feedback from the generator network and the preprocessed data source file, outputs image authenticity labels, and then updates the discriminator network parameters.

[0011] Multiple sets of training samples are used as training data for training;

[0012] After training, testing is performed. The trained generative model is used to test the generative network. The input is a vector with multiple random distributions, and the output is virtual aircraft flight parameter data.

[0013] The virtual aircraft flight parameter data is analyzed, classified, and converted into virtual flight parameter data files in various formats.

[0014] According to one embodiment, the virtual flight parameter data files in various formats include files in the formats ".dat", ".bin", ".mdb", ".xml", and ".txt".

[0015] According to one embodiment, the multiple random distributions include a random uniform distribution and a Gaussian distribution.

[0016] According to one embodiment, the various formats of aircraft flight parameter data source files include files in the formats of ".dat", ".bin", ".mdb", ".xml", and ".txt".

[0017] According to one embodiment, the multiple sets of training samples used during the training process specifically include:

[0018] During training, a single training session uses a set of training samples from a data graph as training data; or,

[0019] During training, a single training session uses multiple sets of training samples from a single data graph as training data; or,

[0020] During training, a single training session may use one set of training samples from each of the multiple data graphs as training data; or,

[0021] During training, a single training session selects multiple sets of training samples from each of the various data graphs as training data.

[0022] According to one embodiment, a fully connected neural network or a deconvolutional network composed of multiple deconvolutional layers is used as the generator network structure, and a fully connected layer or a convolutional network composed of multiple convolutional layers is used as the discriminator network structure; the generator network and the discriminator network mutually stimulate each other to update their parameters.

[0023] According to one embodiment, in the preprocessing, the data graph includes: a time parameter data graph (TPG) with time as the horizontal axis and actual parameter values ​​as the vertical axis, and a parameter parameter data graph (PPG) with one type of actual parameter data value as the horizontal axis and another type of actual parameter data value as the vertical axis.

[0024] Compared with existing technologies, this invention provides a method for generating aircraft flight simulation parameter data. It can generate a large amount of parameter data suitable for applications such as fault diagnosis, fault prediction, and trend analysis in maintenance systems, as well as for research in fields related to big data, cloud computing, data mining, and deep learning, using a relatively small number of aircraft flight parameter data source files in various formats. The generated simulated aircraft flight parameter data is characterized by realism, randomness, and comprehensiveness. It solves the problems of current methods where collecting large amounts of reliable and effective data with the required characteristics is time-consuming, labor-intensive, and requires a huge amount of work, and the collection process is difficult to be accurate, complete, timely, effective, and economical. This invention achieves automatic generation, replacing the manual collection of large amounts of reliable data. The final generated aircraft flight simulation parameter data can be used for applications and research in multiple modules of airborne maintenance systems. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of a method for generating aircraft flight simulation parameter data according to an embodiment of the present invention;

[0027] Figure 2 This is a TPG diagram illustrating the time parameter data of an embodiment of the present invention;

[0028] Figure 3 This is a PPG diagram illustrating the parameter data of an embodiment of the present invention;

[0029] Figure 4 (a), (b), and (c) are schematic diagrams of different time units corresponding to one parameter in the TPG graph of an embodiment of the present invention. Detailed Implementation

[0030] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0031] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments, providing a clear and complete description of the technical solutions of the present invention. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0032] like Figure 1 As shown, this embodiment of the invention provides a method for generating aircraft flight simulation parameter data, including:

[0033] Acquire aircraft flight parameter data source files in various formats;

[0034] The acquired data source files are preprocessed according to different formats. The preprocessing includes classifying the data files, analyzing the characteristics of the file data, and transforming the data into data graphs and images in various ways.

[0035] Constructing a generative network and a discriminative network;

[0036] Multiple randomly distributed vectors are input into the generator network, which simultaneously receives feedback from the discriminator network, thereby updating the generator network parameters.

[0037] The transitional aircraft flight parameter data output by the generator network is fed into the discriminator network. The discriminator network simultaneously receives feedback from the generator network and the preprocessed data source file, outputs image authenticity labels, and then updates the discriminator network parameters.

[0038] Multiple sets of training samples are used as training data for training;

[0039] After training, testing is performed. The trained generative model is used to test the generative network. The input is a vector with multiple random distributions, and the output is virtual aircraft flight parameter data.

[0040] The virtual aircraft flight parameter data is analyzed, classified, and converted into virtual flight parameter data files in various formats.

[0041] This application presents a flexible and controllable method for generating aircraft flight simulation parameter data. It addresses the problem of requiring large amounts of data in civil aircraft airborne maintenance systems for fault diagnosis, fault prediction, health assessment, and aircraft maintenance research based on big data, cloud computing, data mining, and deep learning. This data needs to be reliable, accurate, random, and comprehensive. It also overcomes the challenges of collecting large amounts of reliable and effective data with the required characteristics, which is time-consuming, labor-intensive, and requires a huge investment of time. Furthermore, the collection process is difficult to achieve in terms of accuracy, completeness, timeliness, effectiveness, and economy. This method utilizes a certain number of real aircraft flight parameter data source files in various formats, combined with various random noises as input. It integrates a generator network composed of fully connected neural networks or deconvolutional networks with multiple deconvolutional layers, and a discriminator network composed of fully connected layers or multiple convolutional layers. The method analyzes, classifies, and converts the various formats of aircraft flight parameter data source files into various data graph forms. Then, training data, specifically divided into multiple groups according to different types of data graphs and data characteristics, is used as part of the input to the discriminator network. Simultaneously, various random distribution vectors, such as uniform or Gaussian distributions, are used as generator data. The generator network takes virtual flight parameter data as input and outputs it as output to the discriminator network. The discriminator network then outputs labels indicating the authenticity of the images. During training, the generator and discriminator networks mutually incentivize each other to update their parameters, improving their capabilities. After training, various randomly distributed vectors are used as input to the generator network. Combined with the trained generator model, the generator network outputs virtual aircraft flight parameter data. This data is then converted into various file formats to meet different needs, enabling its use in maintenance systems for fault diagnosis, fault prediction, health assessment, and aircraft maintenance research based on big data, cloud computing, data mining, and deep learning.

[0042] This invention establishes a flexible and controllable method for generating aircraft flight simulation parameter data for civil aircraft airborne maintenance systems. It utilizes various formats of aircraft flight parameter data source files and a generator network composed of deconvolutional networks and a discriminator network to complete the simulation data generation process. This generates various simulated aircraft flight parameter data files used in the maintenance system for fault diagnosis, fault prediction, health assessment, and aircraft maintenance research processes based on big data, cloud computing, data mining, and deep learning. For example... Figure 1 As shown, the specific implementation of the present invention is as follows:

[0043] a. Obtain aircraft flight parameter data source files in various formats, such as ".dat", ".bin", "mdb", ".xml", ".txt", etc.;

[0044] b. Preprocess the acquired data source files according to different formats, including classifying the data files, analyzing the characteristics of the file data, and transforming the data into various data graph and image formats in multiple ways;

[0045] c. There are two types of data charts: one is a time parameter data chart (TPG) with time as the horizontal axis and actual parameter values ​​as the vertical axis, and the other is a parameter parameter data chart (PPG) with actual values ​​of one type of parameter data as the horizontal axis and actual values ​​of another type of parameter data as the vertical axis.

[0046] d. Time parameter data plots and parameter parameter data plots: The horizontal and vertical axes of the graph are specifically represented by two predefined colors. The corresponding parameter value curves in the graph also have different colors. For example, a TPG plot... Figure 2 As shown, the PPG chart is as follows: Figure 3 As shown;

[0047] e. Simultaneously, different colors are used to represent the time axis (T-axis) in the TPG image for different units of time, such as... Figure 3 As shown;

[0048] f. For TPG graphs, multiple sets of training samples can be constructed by using different time units on the horizontal axis and different parameters on the corresponding vertical axis.

[0049] g. For PPG plots, different parameters can be selected based on the horizontal axis and the corresponding vertical axis to form multiple sets of training samples;

[0050] h. Use a randomly generated vector as the input to the generator network. Here, the random vector can generally adopt a common uniform distribution (such as U(0,1)) or a Gaussian distribution (such as N(0,1)).

[0051] i. The generative network can be a fully connected neural network or a deconvolutional network consisting of multiple deconvolutional layers;

[0052] j. The output of the generator network is transitional virtual flight parameter data;

[0053] k. The input to the discriminant network is real flight parameter data and transitional virtual aircraft flight parameter data;

[0054] l. The discriminant network can be a convolutional network consisting of fully connected layers or multiple convolutional layers;

[0055] m. The output of the discriminant network is a label indicating the authenticity of the image;

[0056] n. Training data can be divided into multiple groups:

[0057] o. During training, a single training session can use a set of training samples from a data graph as training data;

[0058] p. During the training process, a single training session can use multiple sets of training samples from a single data graph as training data;

[0059] q. During training, a single training session can use multiple data graphs, from which one set of training samples is selected as training data for each data graph.

[0060] r. During training, a single training session can use multiple data graphs, from which multiple sets of training samples are selected as training data for each data graph.

[0061] s. During training, for a TPG image, the same parameter can correspond to different units of time, thus forming different training samples as a set of training data. This is because different units of time may lead to different trends in the parameter values ​​of the same image size. Figure 4 As shown in (a)(b)(c);

[0062] During training, the generator network and the discriminator network mutually stimulate each other to update their parameters, improving the generator network's ability to generate transitional virtual data. As a result, the transitional virtual aircraft flight parameter data generated in subsequent testing or generation processes is more in line with mission requirements.

[0063] u. After training, test the generator network using the trained generator model. The input is a uniform distribution (e.g., U(0,1)) or a Gaussian distribution (e.g., N(0,1)), and the output is simulated aircraft flight parameter data.

[0064] v. It can analyze and classify virtual aircraft flight parameter data according to the target file format requirements, and convert it into simulated flight parameter data files in various formats such as ".dat", ".bin", ".mdb", ".xml", and ".txt".

[0065] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating aircraft flight simulation parameter data, characterized in that, include: Acquire aircraft flight parameter data source files in various formats; The acquired data source files are preprocessed according to different formats. The preprocessing includes classifying the data files, analyzing the characteristics of the file data, and transforming the data into data graphs and images in various ways. Constructing a generative network and a discriminative network; Multiple randomly distributed vectors are input into the generator network, which simultaneously receives feedback from the discriminator network, thereby updating the generator network parameters. The transitional aircraft flight parameter data output by the generator network is fed into the discriminator network. The discriminator network simultaneously receives feedback from the generator network and the preprocessed data source file, outputs image authenticity labels, and then updates the discriminator network parameters. Multiple sets of training samples are used as training data for training; After training, testing is performed. The trained generative model is used to test the generative network. The input is a vector with multiple random distributions, and the output is virtual aircraft flight parameter data. The virtual aircraft flight parameter data is analyzed, classified, and converted into virtual flight parameter data files in various formats; The virtual flight parameter data files in various formats include files in the formats ".dat", ".bin", ".mdb", ".xml", and ".txt"; the aircraft flight parameter data source files in various formats include files in the formats ".dat", ".bin", ".mdb", ".xml", and ".txt". The generator network structure is a fully connected neural network or a deconvolutional network composed of multiple deconvolutional layers, and the discriminator network structure is a fully connected network or a convolutional network composed of multiple convolutional layers; the generator network and the discriminator network mutually stimulate each other to update their parameters. In the preprocessing, the data graphs include: a time parameter data graph (TPG) with time as the horizontal axis and actual parameter values ​​as the vertical axis, and a parameter parameter data graph (PPG) with actual parameter data values ​​of one type as the horizontal axis and actual parameter data values ​​of another type as the vertical axis.

2. The method according to claim 1, characterized in that, The various random distributions mentioned include random uniform distribution and Gaussian distribution.

3. The method according to claim 1, characterized in that, The multiple training samples used during training specifically include: During training, a single training session uses a set of training samples from a data graph as training data; or, During training, a single training session uses multiple sets of training samples from a single data graph as training data; or, During training, a single training session may use one set of training samples from each of the multiple data graphs as training data; or, During training, a single training session selects multiple sets of training samples from each of the various data graphs as training data.

Images