Aero-engine gas path multi-combination fault data enhancement method and system

Abstract

The invention relates to the technical field of aero-engines, and particularly provides an aero-engine gas path multi-combination fault data enhancement method and system, and the method comprises the steps: obtaining the original data of a gas path fault of an aero-engine, and dividing the original data into a plurality of groups of fault data according to the fault type; capturing time sequence characteristics of the fault data by using a TimeGAN model, and generating extended data based on the time sequence characteristics; mapping the fault data and the corresponding extended data to a potential space, and screening out a high-density data region from the potential space through KDE; and establishing a minimum hyper-sphere based on the high-density data region by using support vector data description, and extracting extended data in the minimum hyper-sphere to obtain optimal extended data. According to the invention, a large amount of fault data consistent with actual gas path multi-combination fault data distribution can be generated.

Description

Technical Field

[0001] The invention belongs to the technical field of aero-engines, and in particular relates to a method and system for enhancing data of multiple combined faults in gas paths of an aero-engine. Background Art

[0002] In recent years, with the rapid development of machine learning, data-driven methods have been widely used in the field of aircraft engine gas path fault diagnosis. However, in practical applications, the probability of gas path faults, especially multiple combination faults, is low, resulting in a scarcity of fault samples, which poses a challenge to fault diagnosis. Using a deep learning model with limited fault samples is prone to overfitting, thereby reducing the accuracy of fault diagnosis.

[0003] The GAN-based data enhancement method can generate new samples that are closer and closer to the real data distribution through adversarial training, and at the same time train the discriminator to distinguish between real and generated data, so it has significant advantages in generating data. By processing fault data with the GAN-based data enhancement method, the amount of training samples for machine learning can be increased.

[0004] However, these methods find it difficult to learn data features when faced with small samples, non-balanced and highly coupled complex fault raw data, and the data finally generated does not fit the features of the original data very well. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method and system for enhancing data of multiple combination faults in air path of an aircraft engine to solve the above-mentioned technical problems.

[0006] In a first aspect, the present invention provides a method for enhancing data of multiple combined faults in an aircraft engine gas path, comprising: Acquire original data of gas path faults of the aircraft engine, and divide the original data into multiple groups of fault data according to fault categories; Using the TimeGAN model to capture the time series characteristics of the fault data, and generating extended data based on the time series characteristics; Mapping the fault data and the corresponding extended data to a latent space, and filtering out a high-density data region in the latent space through KDE; A minimum hypersphere is established based on the high-density data region using support vector data description, and extended data inside the minimum hypersphere is extracted to obtain optimal extended data.

[0007] In a second aspect, the present invention provides an aircraft engine gas path multiple combination fault data enhancement system, comprising: A data acquisition module, used to acquire raw data of air path faults of aircraft engines, and divide the raw data into multiple groups of fault data according to fault categories; A data extension module, used to capture the time series characteristics of the fault data using a TimeGAN model, and generate extended data based on the time series characteristics; A distribution estimation module, used for mapping the fault data and the corresponding extended data to a latent space, and filtering out a high-density data region in the latent space through KDE; The data screening module is used to establish a minimum hypersphere based on the high-density data area using support vector data description, extract the extended data inside the minimum hypersphere, and obtain the best extended data.

[0008] The beneficial effect of the present invention is that the method and system for enhancing the data of multiple combination faults in the gas path of an aircraft engine provided by the present invention use the TimeGAN-based method for the data enhancement of multiple combination faults in the gas path of an aircraft engine for the first time. The original data of the high-dimensional sensor can be learned through Extraction TimeGAN, and a mechanism for optimal data generation can be established through KDE and SVDD. The quality of the generated data is improved in the comparison of evaluation indicators, and a large amount of fault data consistent with the actual distribution of multiple combination fault data in the gas path can be generated. The present invention effectively solves the problems of small samples and imbalance of multiple combination faults in the gas path of an aircraft engine through Extraction TimeGAN, and improves the performance of the data-driven method. Extraction TimeGAN can generate high-quality multidimensional time series data that is consistent with the distribution of the original data, and can reduce the generation of abnormal data, establish a bridge between the generated data and the application of gas path multiple combination fault diagnosis, so that a simpler model can also achieve high-precision diagnosis of multiple combination faults in the gas path of an aircraft engine, and improve the feasibility of various intelligent diagnosis models in practical applications.

[0009] In addition, the invention has a reliable design principle, a simple structure and a very broad application prospect. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0011] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention.

[0012] Figure 2 It is a thermal diagram of various air path parameters of an aircraft engine according to a method of an embodiment of the present invention.

[0013] Figure 3It is a TimeGAN model structure diagram of a method according to an embodiment of the present invention.

[0014] Figure 4 FIG. 4 is a diagram of an SVDD algorithm structure of a method according to an embodiment of the present invention.

[0015] Figure 5 This is a rendering comparing the original data of the method of an embodiment of the present invention and the data generated by Extraction TimeGAN.

[0016] Figure 6 is a schematic block diagram of a system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0019] The method for enhancing data of multiple combined faults in air paths of aircraft engines provided by the embodiment of the present invention is executed by a computer device, and accordingly, the system for enhancing data of multiple combined faults in air paths of aircraft engines runs in the computer device.

[0020] Figure 1 is a schematic flow chart of a method according to an embodiment of the present invention. Figure 1 The execution subject may be an aircraft engine gas path multiple combination fault data enhancement system. According to different requirements, the order of the steps in the flow chart may be changed, and some may be omitted.

[0021] like Figure 1 As shown, the method includes: Step 110, obtaining original data of the air path fault of the aircraft engine, and dividing the original data into multiple groups of fault data according to the fault category; Step 120, using the TimeGAN model to capture the time series characteristics of the fault data, and generating extended data based on the time series characteristics; Step 130, mapping the fault data and the corresponding extended data to a latent space, and filtering out a high-density data region in the latent space through KDE; Step 140: using support vector data to describe the establishment of a minimum hypersphere based on the high-density data region, extracting the extended data inside the minimum hypersphere, and obtaining the optimal extended data.

[0022] To facilitate understanding of the present invention, the following further describes the method for enhancing data of multiple combined faults in an aero-engine gas path of the present invention based on the principle of the method and in combination with an embodiment.

[0023] Specifically, the method for enhancing data of multiple combined faults of aero-engine gas path includes: S1. Acquire original data of air path faults of an aircraft engine, and divide the original data into multiple groups of fault data according to fault categories.

[0024] A certain type of low bypass ratio turbofan engine was taken as the research object, and a digital model of the aircraft engine was established in Gasturb software. The aircraft engine mainly consists of an air inlet, a fan, a high-pressure compressor, a high-pressure turbine, an annular combustion chamber, a low-pressure turbine and a tail nozzle.

[0025] Gasturb software is used to inject single faults, double combination faults and triple combination faults into the fan, HPC, HPT and LPT respectively. The fault types include blade fouling, blade wear, blade corrosion, blade mechanical damage, etc. The gas path multiple combination fault settings are shown in Table 1. In the table, f represents flow, e represents efficiency, and the subscript represents the fault location. The flight parameters selected for aircraft engine gas path fault diagnosis include high-pressure rotor speed, low-pressure rotor speed, total fuel flow, total inlet temperature, total inlet pressure, etc.

[0026] Aircraft engine gas path parameters such as temperature, pressure, flow, etc. affect each other. The change of one parameter will cause a chain reaction of other parameters. In order to analyze the correlation between the various air path parameters of aircraft engines, the thermal diagram of each air path parameter is given as follows Figure 2As shown. As can be seen from the figure, the present invention selects 12 aircraft engine gas path parameters to form a total of 144 pairs of correlations. The correlation coefficient with an absolute value of more than 0.7 is regarded as a strong correlation, the correlation coefficient between 0.3 and 0.7 is regarded as a medium correlation, and the correlation coefficient less than 0.3 is regarded as a weak correlation. According to statistics, there are 32 strong correlations, accounting for 22.22%, 29 medium correlation coefficients, accounting for 20.14%, and 83 weak correlations, accounting for 57.64%. Therefore, the data generated by the air path of the aircraft engine needs to pay attention to the correlation and coupling between multiple combination faults and between sensor parameters at the same time, which puts higher requirements on the performance of the data enhancement method.

[0027] Inject Gaussian white noise with SNR20 into the original data.

[0028] S2. Use the TimeGAN model to capture the time series characteristics of the fault data, and generate extended data based on the time series characteristics.

[0029] Please refer to Figure 3 TimeGAN is mainly composed of four parts: embedding network, recurrence network, generator and discriminator. The embedding network is an autoencoder that can map high-dimensional data to latent space, mining and capturing high-value feature information in high-dimensional vectors; the recurrence network restores the latent space to the initial dimension, minimizing the data reconstruction loss. The generative adversarial network trains the generator and discriminator in the embedding space, which can fit the temporal correlation of high-dimensional data and effectively reduce the complexity of adversarial model training, thereby achieving better learning and training effects.

[0030] Initialize a set of features With joint distribution p TimeGAN decomposes the time series data features into static features and time series features. Let vector Indicates that there is Static characteristics of aircraft engine gas path data; vector Indicates that there is The time series characteristics of the air path data of an aircraft engine; and They are the vector spaces of static and time series characteristics of aircraft engine gas path data; The distribution of the original data is defined as , the distribution of the extended data is defined as ;in, T is the length of its time series; let , represents the index of a single sample in the training data, then the training data set is represented as ; Time stepwise conditional distribution obtained by autoregressive decomposition , the training objective is decomposed into the overall distribution learning objective of aircraft engine gas path data and the time-dependent distribution learning objective: The overall distribution learning objectives of aircraft engine gas path data are: , In the formula, represents the distribution distance; The learning objective of the time-dependent distribution of aircraft engine gas path data is: ; The original data of the air path of the aircraft engine is reconstructed by using the embedding function and the recurrence function to obtain the mapping between the air path data features and the latent space. and The static feature space of aero-engine gas path data and time series feature space The corresponding latent feature space has an embedding function: ; The static and temporal features of the original data of the aero-engine gas path are converted into latent space encoding through the embedding function and : , where Embedding network for static features of raw data of air path of aircraft engine; Embedding network for the time series characteristics of raw data of aircraft engine gas path; Reproduce the network Reconstruct the static and time series features of aircraft engine gas path data into ,Right now: , where A reproduction network for the static characteristics of the original data of the air path of an aircraft engine; A reproduction network for the time series characteristics of the original data of the aircraft engine gas path; Define a vector space with known distribution and , generating random vectors as distribution and , then the generating function is , the potential codes for generating static features and temporal features of the original data of the aircraft engine gas path are: , where Generate a network of static features for generating data for aero-engine gas paths; A network for generating time series features of data for aero-engine gas paths; Discriminant function Receive static features and time series features of aircraft engine gas path generation data and output classification results: , where is the forward hidden state sequence; is the reverse hidden state sequence; is a forward recursive function; is a backward recursive function; Generate static feature discriminant function for aircraft engine gas path data; Generate a discriminant function for the time series characteristics of data for the air path of an aircraft engine; The TimeGAN model uses the reconstruction loss of the embedding network and the recurrence network Optimize the encoding and decoding of the autoencoder; introduce the original data of the air path of the aircraft engine as a supervision item, through the loss between the generator and the original data Quantitatively judge the learning ability and data quality of the generator; finally define the adversarial loss , to achieve feedback correction of the generator; the loss functions of the TimeGAN model include: ; ; ; TimeGAN minimizes , and Three loss functions capture the spatiotemporal relationships and structures in time series data, thereby generating extended data with similar temporal characteristics.

[0031] S3. Mapping the fault data and the corresponding extended data to a latent space, and filtering out a high-density data region in the latent space through KDE.

[0032] The data generated by TimeGAN does not completely fit the distribution of the original data. In order to improve the quality of the generated data and make the generated data as close to the distribution of the original data as possible, KDE is specially introduced to optimize the generated data that conforms to the distribution of high-density original data.

[0033] Kernel Density Estimation (KDE) is a nonparametric method for estimating probability density functions. and Represent the probability density function and cumulative distribution function of the aircraft engine gas path data, respectively, and extract n Independent and identically distributed data , whose sample value is , then in x KDE value of the distribution function at for: , where is the selected KDE kernel function, h is bandwidth; The kernel function is a Gaussian kernel function, that is: ; The bandwidth uses the optimal fixed bandwidth ; Use integrated mean square error to evaluate the estimated probability density distribution and the true probability density distribution Error ,Right now: , In the formula, is the integrated mean square error value; is the mathematical expectation; is the asymptotic integrated mean square error value; is a high-order term; After discarding the high-order terms, the asymptotic integrated mean square error is: ; The asymptotic integrated mean square error is h The derivative of , and set the derivative equal to 0, solve for the optimal fixed bandwidth : ; Restricted to the true probability density function Unknown, determine the optimal fixed bandwidth through cross-validation ; Through Integrate to calculate the cumulative distribution function CDF : ; Set cumulative distribution threshold , select the cumulative distribution function to exceed the cumulative distribution threshold All data are obtained to obtain the KDE high-density data area: .

[0034] KDE uses a statistical method to determine the high-density data area. If the generated data of TimeGAN is located in the high-density area of ​​KDE, it is closer to the distribution of the original data and the quality of the generated data is higher. Otherwise, it is considered that the generated data has a certain deviation from the distribution of the original data and the quality of the generated data is relatively low. Therefore, in the following text, a hypersphere based on the high-density original data of KDE is constructed through SVDD, and the generated data inside the hypersphere is selected and saved as the preferred generated data.

[0035] S4. Using support vector data to describe the establishment of a minimum hypersphere based on the high-density data region, extracting the extended data inside the minimum hypersphere, and obtaining the optimal extended data.

[0036] like Figure 4As shown in the figure, Support Vector Data Description (SVDD) is a type of method used for anomaly detection. The KDE high-density original data is mapped to a high-dimensional space through a kernel function, and a minimum hypersphere is established to contain all input samples. The samples on the minimum hypersphere are called support vectors. When the generated data is inside the hypersphere, it is identified as generated data that conforms to the distribution of the high-density original data, and is also called the preferred generated data. If it is outside the hypersphere, it is identified as generated data that is not distributed in the high-density original data. The preferred generated data part of the generated data located inside the hypersphere is extracted, and the data located outside the hypersphere is eliminated, so as to improve the quality of the generated data.

[0037] The original data sequence of the KDE high-density data area , through nonlinear mapping to high-dimensional space, in the kernel feature space F In order to make the hypersphere In order to include these core sample sets as much as possible, the constraints that need to be satisfied when constructing the minimized hypersphere include: , In the formula, is a nonlinear mapping function; is the center of the hypersphere; is the radius of the hypersphere; is the penalty factor; is the slack variable; By using the Lagrange multiplier method, the dual form of the optimization problem with the constraints is expressed as:

[0038] In the formula, and Respectively i and j training samples; is the kernel function of the inner product in high-dimensional space; is the Lagrange multiplier; The center of the hypersphere after update and radius The calculation formula includes: , , In the formula, is the support vector; Newly input aircraft engine gas path fault data points With the center of the hypersphere Distance for: ; By comparison and Size judgment extended data Is it the best extended data? Then determine the data point On or inside the smallest hypersphere, it is the best extended data; if Then determine the data point Outside the smallest hypersphere, it is not the best expansion data.

[0039] S5. Data quality assessment.

[0040] Comparison of the original gas path data and the data generated by the Extraction TimeGAN model at the same time step Figure 5 As shown in the figure, the horizontal axis is the time step and the vertical axis is the parameter amplitude. From the perspective of short-term change trends, although the changes in the generated data and the original data are not strictly consistent, they have a certain trend and can retain the temporal correlation and transient distribution characteristics of the original data; from the perspective of specific values, the difference in values ​​between the generated data and the original data not only effectively expands the scale of the data set, but also enhances the representativeness and coverage of the data. Therefore, the data enhancement strategy based on Extraction TimeGAN provides a more comprehensive and balanced data foundation for subsequent data analysis and machine learning tasks, which is expected to improve the generalization ability and prediction accuracy of the model.

[0041] In this embodiment, the aircraft engine gas path multiple combination fault data enhancement system can be divided into multiple functional modules according to the functions it performs, such as Figure 6 As shown. The functional modules of the system 600 may include: a data acquisition module 610, a data expansion module 620, a distribution estimation module 630 and a data screening module 640. The module referred to in the present invention refers to a series of computer program segments that can be executed by at least one processor and can complete fixed functions, which are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0042] A data acquisition module, used to acquire raw data of air path faults of aircraft engines, and divide the raw data into multiple groups of fault data according to fault categories; A data extension module, used to capture the time series characteristics of the fault data using a TimeGAN model, and generate extended data based on the time series characteristics; A distribution estimation module, used for mapping the fault data and the corresponding extended data to a latent space, and filtering out a high-density data region in the latent space through KDE; The data screening module is used to establish a minimum hypersphere based on the high-density data area using support vector data description, extract the extended data inside the minimum hypersphere, and obtain the best extended data.

[0043] Although the present invention has been described in detail with reference to the accompanying drawings and in combination with the preferred embodiments, the present invention is not limited thereto. Without departing from the spirit and essence of the present invention, a person of ordinary skill in the art may make various equivalent modifications or substitutions to the embodiments of the present invention, and these modifications or substitutions shall be within the scope of the present invention. Any person of ordinary skill in the art may easily think of changes or substitutions within the technical scope disclosed by the present invention, and these shall be within the scope of protection of the present invention.

Claims

1. A method for enhancing data of multiple combined faults in the gas path of an aircraft engine, characterized in that: include: Acquire original data of gas path faults of the aircraft engine, and divide the original data into multiple groups of fault data according to fault categories; Using the TimeGAN model to capture the time series characteristics of the fault data, and generating extended data based on the time series characteristics; Mapping the fault data and the corresponding extended data to a latent space, and filtering out a high-density data region in the latent space through KDE; A minimum hypersphere is established based on the high-density data region using support vector data description, and extended data inside the minimum hypersphere is extracted to obtain optimal extended data.

2. The method according to claim 1, characterized in that The original data of the air path fault of the aircraft engine is obtained, and the original data is divided into multiple groups of fault data according to the fault category, including: Gasturb software was used to inject single faults, double combination faults, and triple combination faults into the fan, HPC, HPT, and LPT, respectively; Collecting fault diagnosis related parameters, wherein the fault diagnosis related parameters include high pressure rotor speed, low pressure rotor speed, total fuel flow, total air inlet temperature, and total air inlet pressure; Collect aircraft engine gas path parameters, including temperature, pressure, and flow; The injected faults and corresponding fault diagnosis related parameters and aircraft engine gas path parameters are saved in the original data repository; Inject Gaussian white noise with SNR20 into the original data.

3. The method according to claim 1, characterized in that The TimeGAN model is used to capture the time series characteristics of the fault data, and extended data is generated based on the time series characteristics, including: Initialize a set of features With joint distribution p ,vector Indicates that there is Static characteristics of aircraft engine gas path data; vector Indicates that there is The time series characteristics of the air path data of an aircraft engine; and They are the vector spaces of static and time series characteristics of aircraft engine gas path data; The distribution of the original data is defined as , the distribution of the extended data is defined as ;in, T is the length of its time series; let , represents the index of a single sample in the training data, then the training data set is represented as ; Time stepwise conditional distribution obtained by autoregressive decomposition , the training objective is decomposed into the overall distribution learning objective of aircraft engine gas path data and the time-dependent distribution learning objective: The overall distribution learning objectives of aircraft engine gas path data are: , In the formula, represents the distribution distance; The learning objective of the time-dependent distribution of aircraft engine gas path data is: ; The original data of the air path of the aircraft engine is reconstructed by using the embedding function and the recurrence function to obtain the mapping between the air path data features and the latent space. and The static feature space of aero-engine gas path data and time series feature space The corresponding latent feature space has an embedding function: ; The static and temporal features of the original data of the aero-engine gas path are converted into latent space encoding through the embedding function and : , where Embedding network for static features of raw data of air path of aircraft engine; Embedding network for the time series characteristics of raw data of aircraft engine gas path; Reproduce the network Reconstruct the static and time series features of aircraft engine gas path data into ,Right now: , where A reproduction network for the static characteristics of the original data of the air path of an aircraft engine; A reproduction network for the time series characteristics of the original data of the aircraft engine gas path; Define a vector space with known distribution and , generating random vectors as distribution and , then the generating function is , the potential codes for generating static features and temporal features of the original data of the aircraft engine gas path are: , where Generate a network of static features for generating data for aero-engine gas paths; A network for generating time series features of data for aero-engine gas paths; Discriminant function Receive static features and time series features of aircraft engine gas path generation data and output classification results: , where is the forward hidden state sequence; is the reverse hidden state sequence; is a forward recursive function; is a backward recursive function; Generate static feature discriminant function for aircraft engine gas path data; Generate a discriminant function for the time series characteristics of data for the air path of an aircraft engine; The TimeGAN model uses the reconstruction loss of the embedding network and the recurrence network Optimize the encoding and decoding of the autoencoder; introduce the original data of the air path of the aircraft engine as a supervision item, through the loss between the generator and the original data Quantitatively judge the learning ability and data quality of the generator; finally define the adversarial loss , to achieve feedback correction of the generator; the loss functions of the TimeGAN model include: ； ； ； TimeGAN minimizes , and Three loss functions capture the spatiotemporal relationships and structures in time series data, thereby generating extended data with similar temporal characteristics.

4. The method according to claim 3, characterized in that The fault data and the corresponding extended data are mapped to a latent space, and a high-density data area is screened out from the latent space through KDE, including: use and Represent the probability density function and cumulative distribution function of the aircraft engine gas path data, respectively, and extract n Independent and identically distributed data , whose sample value is , then in x The KDE value of the distribution function at for: , where is the selected KDE kernel function, h is bandwidth; The kernel function is a Gaussian kernel function, that is: ; The bandwidth uses the optimal fixed bandwidth ; Use integrated mean square error to evaluate the estimated probability density distribution and the true probability density distribution Error ,Right now: ， In the formula, is the integrated mean square error value; is the mathematical expectation; is the asymptotic integrated mean square error value; is a high-order term; After discarding the high-order terms, the asymptotic integrated mean square error is: ; The asymptotic integrated mean square error is h The derivative of , and set the derivative equal to 0, solve for the optimal fixed bandwidth : ； Restricted to the true probability density function Unknown, determine the optimal fixed bandwidth through cross-validation ; Through Integrate to calculate the cumulative distribution function CDF : ； Set cumulative distribution threshold , select the cumulative distribution function to exceed the cumulative distribution threshold All data are obtained to obtain the KDE high-density data area: .

5. The method according to claim 4, characterized in that Using support vector data to describe the establishment of a minimum hypersphere based on the high-density data area, extracting extended data inside the minimum hypersphere, and obtaining optimal extended data, including: The original data sequence of the KDE high-density data area , through nonlinear mapping to high-dimensional space, in the kernel feature space F In order to make the hypersphere In order to include these core sample sets as much as possible, the constraints that need to be satisfied when constructing the minimized hypersphere include: ， In the formula, is a nonlinear mapping function; is the center of the hypersphere; is the radius of the hypersphere; is the penalty factor; is the slack variable; By using the Lagrange multiplier method, the dual form of the optimization problem with the constraints is expressed as: In the formula, and Respectively i and j training samples; is the kernel function of the inner product in high-dimensional space; is the Lagrange multiplier; The center of the hypersphere after update and radius The calculation formula includes: ， ， In the formula, is the support vector; Newly input aircraft engine gas path fault data points With the center of the hypersphere Distance for: ； By comparison and Size judgment extended data Is it the best extended data? Then determine the data point On or inside the smallest hypersphere, it is the best extended data; if Then determine the data point Outside the smallest hypersphere, it is not the best expansion data.

6. An aircraft engine gas path multiple combination fault data enhancement system, characterized in that: include: A data acquisition module, used to acquire raw data of air path faults of aircraft engines, and divide the raw data into multiple groups of fault data according to fault categories; A data extension module, used to capture the time series characteristics of the fault data using a TimeGAN model, and generate extended data based on the time series characteristics; A distribution estimation module, used for mapping the fault data and the corresponding extended data to a latent space, and filtering out a high-density data region in the latent space through KDE; The data screening module is used to establish a minimum hypersphere based on the high-density data area using support vector data description, extract the extended data inside the minimum hypersphere, and obtain the best extended data.

7. The system according to claim 6, characterized in that: The data acquisition module is used for: Gasturb software was used to inject single faults, double combination faults, and triple combination faults into the fan, HPC, HPT, and LPT, respectively; Collecting fault diagnosis related parameters, wherein the fault diagnosis related parameters include high pressure rotor speed, low pressure rotor speed, total fuel flow, total air inlet temperature, and total air inlet pressure; Collect aircraft engine gas path parameters, including temperature, pressure, and flow; The injected faults and corresponding fault diagnosis related parameters and aircraft engine gas path parameters are saved in the original data repository; Inject Gaussian white noise with SNR20 into the original data.

8. The system according to claim 6, characterized in that The data expansion module is used for: Initialize a set of features With joint distribution p ,vector Indicates that there is Static characteristics of aircraft engine gas path data; vector Indicates that there is The time series characteristics of the air path data of an aircraft engine; and They are the vector spaces of static and time series characteristics of aircraft engine gas path data; The distribution of the original data is defined as , the distribution of the extended data is defined as ;in, T is the length of its time series; let , represents the index of a single sample in the training data, then the training data set is represented as ; Time stepwise conditional distribution obtained by autoregressive decomposition , the training objective is decomposed into the overall distribution learning objective of aircraft engine gas path data and the time-dependent distribution learning objective: The overall distribution learning objectives of aircraft engine gas path data are: , In the formula, represents the distribution distance; The learning objective of the time-dependent distribution of aircraft engine gas path data is: ; The original data of the air path of the aircraft engine is reconstructed by using the embedding function and the recurrence function to obtain the mapping between the air path data features and the latent space. and The static feature space of aero-engine gas path data and time series feature space The corresponding latent feature space has an embedding function: ; The static and temporal features of the original data of the aero-engine gas path are converted into latent space encoding through the embedding function and : , where Embedding network for static features of raw data of air path of aircraft engine; Embedding network for the time series characteristics of raw data of aircraft engine gas path; Reproduce the network Reconstruct the static and time series features of aircraft engine gas path data into ,Right now: , where A reproduction network for the static characteristics of the original data of the air path of an aircraft engine; A reproduction network for the time series characteristics of the original data of the aircraft engine gas path; Define a vector space with known distribution and , generating random vectors as distribution and , then the generating function is , the potential codes for generating static features and temporal features of the original data of the aircraft engine gas path are: , where Generate a network of static features for generating data for aero-engine gas paths; A network for generating time series features of data for aero-engine gas paths; Discriminant function Receive static features and time series features of aircraft engine gas path generation data and output classification results: , where is the forward hidden state sequence; is the reverse hidden state sequence; is a forward recursive function; is a backward recursive function; Generate static feature discriminant function for aircraft engine gas path data; Generate a discriminant function for the time series characteristics of data for the air path of an aircraft engine; The TimeGAN model uses the reconstruction loss of the embedding network and the recurrence network Optimize the encoding and decoding of the autoencoder; introduce the original data of the air path of the aircraft engine as a supervision item, through the loss between the generator and the original data Quantitatively judge the learning ability and data quality of the generator; finally define the adversarial loss , to achieve feedback correction of the generator; the loss functions of the TimeGAN model include: ； ； ； TimeGAN minimizes , and Three loss functions capture the spatiotemporal relationships and structures in time series data, thereby generating extended data with similar temporal characteristics.

9. The system according to claim 8, characterized in that The distribution estimation module is used to: use and Represent the probability density function and cumulative distribution function of the aircraft engine gas path data, respectively, and extract n Independent and identically distributed data , whose sample value is , then in x The KDE value of the distribution function at for: , where is the selected KDE kernel function, h is bandwidth; The kernel function is a Gaussian kernel function, that is: ; The bandwidth uses the optimal fixed bandwidth ; Use integrated mean square error to evaluate the estimated probability density distribution and the true probability density distribution Error ,Right now: ， In the formula, is the integrated mean square error value; is the mathematical expectation; is the asymptotic integrated mean square error value; is a high-order term; After discarding the high-order terms, the asymptotic integrated mean square error is: ; The asymptotic integrated mean square error is h The derivative of , and set the derivative equal to 0, solve for the optimal fixed bandwidth : ； Restricted to the true probability density function Unknown, determine the optimal fixed bandwidth through cross-validation ; Through Integrate to calculate the cumulative distribution function CDF : ； Set cumulative distribution threshold , select the cumulative distribution function to exceed the cumulative distribution threshold All data are obtained to obtain the KDE high-density data area: .

10. The system according to claim 9, characterized in that The data screening module is used for: The original data sequence of the KDE high-density data area , through nonlinear mapping to high-dimensional space, in the kernel feature space F In order to make the hypersphere In order to include these core sample sets as much as possible, the constraints that need to be satisfied when constructing the minimized hypersphere include: ， In the formula, is a nonlinear mapping function; is the center of the hypersphere; is the radius of the hypersphere; is the penalty factor; is the slack variable; By using the Lagrange multiplier method, the dual form of the optimization problem with the constraints is expressed as: In the formula, and Respectively i and j training samples; is the kernel function of the inner product in high-dimensional space; is the Lagrange multiplier; The center of the hypersphere after update and radius The calculation formula includes: ， ， In the formula, is the support vector; Newly input aircraft engine gas path fault data points With the center of the hypersphere Distance for: ； By comparison and Size judgment extended data Is it the best extended data? Then determine the data point On or inside the smallest hypersphere, it is the best extended data; if Then determine the data point Outside the smallest hypersphere, it is not the best expansion data.

Images