Aero-engine fault prediction method, system, device and storage medium
By using minimum/maximum autocorrelation analysis and support vector regression models, the problem of not considering time correlation in aero-engine failure prediction is solved, and more accurate prediction of remaining service life is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHENGDU UNIV OF INFORMATION TECH
- Filing Date
- 2026-02-04
- Publication Date
- 2026-05-29
AI Technical Summary
Existing methods for predicting aero-engine failures fail to effectively consider time-related factors, resulting in low accuracy in predicting remaining service life.
The Minimum/Maximum Autocorrelation Analysis (MAF) method was used to extract strong autocorrelation factors, which were then combined with a Support Vector Regression (SVR) model to construct a training set and perform online prediction.
It improves the accuracy of predicting the remaining service life of aero engines and can better reflect the dynamic characteristics of the system degradation process.
Smart Images
Figure CN122113045A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of equipment failure prediction technology, and particularly relates to methods, systems, equipment and storage media for predicting failures of aero-engines. Background Technology
[0002] Aero engines are highly complex aerodynamic and thermodynamic rotating mechanical systems, typically operating under conditions of continuous high temperature, high pressure, high-speed rotation, and intense vibration. Consequently, critical components of aero engines continuously degrade. After a certain degree of component degradation, the performance level of the aero engine will deviate from its normal operating condition, potentially leading to various failures. A failure can cause enormous economic losses and casualties; therefore, monitoring the engine's health status and proactively maintaining and servicing it is of paramount importance. Failure prediction, by monitoring the system's operating status and predicting the time of failure or remaining service life, can significantly reduce equipment or system maintenance costs and improve reliability and safety. Failure prediction is a crucial aspect of equipment health management, involving the estimation of the failure time and risk of one or more existing or future failure modes. Reliable failure prediction is invaluable information, providing maintenance personnel with time to take action and helping to prevent catastrophic situations.
[0003] However, predicting the failure time (or remaining service life) of aero-engines is a challenging task. Currently, methods for predicting failure time (or remaining service life) mainly fall into two categories: 1) model-based methods and 2) data-driven methods. Model-based methods are more accurate when the system is relatively simple and its degradation mechanism is easy to analyze and model, but they require prior knowledge of the physical system, which is often difficult to obtain in practice. Data-driven methods require less prior knowledge, mainly relying on historical degradation trajectory data obtained from sensors. They use machine learning methods to reveal the underlying causal relationships, thereby inferring the remaining service life. However, existing data-driven RUL prediction methods do not consider time correlation when building degradation models, thus failing to accurately reflect the dynamic characteristics of degradation, resulting in low accuracy in predicting remaining service life. Summary of the Invention
[0004] The purpose of this invention is to provide a method, system, device and storage medium for predicting aircraft engine failures, so as to solve the problem that the existing technology does not consider time correlation when establishing degradation models and has low accuracy in predicting the remaining service life.
[0005] The embodiment of this application is implemented as follows: the aircraft engine fault prediction method includes: S01. Obtain the characteristic variables; S02. Based on the characteristic variables, use the MAF method to perform minimum / maximum autocorrelation factor analysis and extract strong autocorrelation factors; S03. Construct a training set using the extracted strong autocorrelation factors and their corresponding remaining lifetimes; S04. Obtain a support vector regression model using the training set; S05. Use support vector regression models to predict the remaining service life of aero engines online.
[0006] Optionally, in some embodiments of this application, characteristic variables are obtained through wrap-around variable selection, wherein the characteristic variables include the total outlet temperature of the high-pressure compressor, denoted as... The total outlet temperature of the low-pressure turbine is set to... The fan inlet pressure is set to... .
[0007] Optionally, in some embodiments of this application, the method for extracting strong autocorrelation factors includes the following steps: S021. Let the original space variable be... In the formula, This refers to the total outlet temperature of the high-pressure compressor. This refers to the total outlet temperature of the low-pressure turbine. Let t be the fan inlet pressure and t be the current time; let the time lag parameter be... For the first A sample matrix consisting of time series The sample matrix consists of rows representing samples and columns representing state variables. A time lag parameter is defined. ; S022, For Find the state difference matrix of the aero-engine. The formula is as follows: ; In the formula, t represents the current time. For time lag parameters, For the first A time series, For the first A sample matrix consisting of time series; S023, Calculate the variance-covariance matrix and ,as follows: ; ; In the formula, t represents the current time. For time lag parameters, For the first A time series, For the first A sample matrix consisting of n time series, where n is the number of samples in the sample matrix, and T is the transpose of the matrix.
[0008] S024, Calculate the matrix * ,in, express The false reversal; S025. Solve the following eigenvalue decomposition problem: ( * ) = ; In the formula, The covariance matrix is the sample lag time with zero. Lag time The covariance matrix, for The false rebellion, For the first A time series, Let λ be the eigenvector and λ be the eigenvalue.
[0009] Optionally, in some embodiments of this application, the following eigenvalue decomposition problem is solved, including: S0251, The eigenvalues are arranged in ascending order, that is: The eigenvectors corresponding to these eigenvalues are the projection vectors of the autocorrelation factors, respectively. Let the projection vector be... When, i.e., the projection vector is For the original space sample The autocorrelation factor projected onto the projection vector The autocorrelation coefficient of this autocorrelation factor Obviously there is ; S0252. Arrange the correlation coefficients from largest to smallest, and set them as follows: The corresponding projection vectors are respectively Given a threshold ,like Then the former Factors , , ..., It exhibits relatively strong autocorrelation, reflecting the autocorrelation characteristics of the system process state, and is called a strong autocorrelation factor; later Factors with weak autocorrelation reflect noise in the system and are called weak autocorrelation factors.
[0010] Optionally, in some embodiments of this application, strong autocorrelation factors are extracted to obtain the projection matrix of each strong autocorrelation factor. ; ; For a new sample representing the state of an aircraft engine , Strong autocorrelation factor It can be calculated using the following transformation formula: .
[0011] Optionally, in some embodiments of this application, the method for constructing the training set includes: S031, Let the first A data matrix consisting of time series , , The data matrix of the time series, composed of strong autocorrelation components, was obtained after analysis using the MAF method. , , The time series is decomposed into non-overlapping windows of size L, where L is a parameter; The Each window is composed of Composition; from each window, extract two parameters for each dimension: the average value within the i-th window. i and the linear regression trend coefficient on the window i Therefore, each window can be calculated to have a size of eigenvectors, where This represents the number of strongly autocorrelated factors in the MAF analysis; S032. Associate the remaining lifetime with each feature vector: For The Each window has a corresponding remaining lifespan calculated as follows: For the entire time series set Repeat the above steps to obtain the training set. ,in This is a vector composed of the features of each hadron correlation factor extracted within each window. It represents the remaining lifetime of the time series starting from the last moment of the corresponding window.
[0012] Optionally, in some embodiments of this application, a support vector regression (SVR) model is used to predict the remaining service life of the aircraft engine online, including the following steps: S051. Assume the aircraft engine testing process is a time series. Let the current time be the last monitoring time. ; S052, Time Series After variable screening, MAF analysis generates time series containing strongly autocorrelated variables. For the current time point, we take... - L to The strongly autocorrelated variable sequences are used as the time window, where the window length is set to L=20, and the window feature vector is extracted and set as follows. ,Will Applied to the Support Vector Regression (SVR) model, it obtains the remaining service life of the aircraft engine under the current monitoring status, i.e.: ,in This is a support vector regression model.
[0013] Accordingly, embodiments of this application also provide an aircraft engine fault prediction system, including: The feature variable module is used to obtain feature variables; The strong autocorrelation factor module uses the MAF method to perform minimum / maximum autocorrelation factor analysis based on characteristic variables to extract strong autocorrelation factors. The training set construction module utilizes the extracted strong autocorrelation factors and their corresponding remaining lifetimes to construct the training set; The Support Vector Regression Model module uses the training set to obtain the support vector regression model. The online prediction module uses a support vector regression model to predict the remaining service life of aero engines online.
[0014] Accordingly, embodiments of this application also provide a computer device, including a storage device and a processor, wherein the storage device stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.
[0015] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.
[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This application removes noise by performing minimum / maximum autocorrelation analysis on sensor data and discarding weak autocorrelation factors. To further improve the stability of the prediction model, an SVR degradation model is established based on the strong autocorrelation factors of MAF. To enhance the stability of the degradation model, the strong autocorrelation factors of the aero-engine during the degradation process are divided into windows, and the mean and slope of the strong autocorrelation factors are extracted as two stable features as independent variables for support vector regression. This model better reflects the dynamic characteristics of the system degradation process and provides better remaining lifetime prediction results. This application introduces minimum-maximum autocorrelation analysis for the first time to analyze the state variables of aero-engines, and then uses support vector regression to establish a support vector regression (SVR) model for the maximum autocorrelation factor. Finally, the SVR model is used to predict the remaining lifetime of the current state, thereby improving the accuracy of the remaining lifetime prediction. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of an aircraft engine structure; Figure 2 This is a flowchart of the aircraft engine fault prediction method of the present invention; Figure 3 This invention provides a strong autocorrelation factor. Window division and feature extraction diagram. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0019] Aero engines typically suffer from slow and long degradation processes, and in practical applications, there is rarely complete operational data from normal operation to failure. The turbofan engine is a widely used type of aero engine, and its structure is illustrated in the diagram below. Figure 1As shown, it mainly consists of a fan, low-pressure compressor, high-pressure compressor, combustion chamber, low-pressure rotor, high-pressure rotor, low-pressure turbine, high-pressure turbine, and exhaust nozzle. NASA Research Center in the United States constructed a simulation model of the degradation process of turbofan aero-engines based on the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS), and generated simulation datasets of different engine models from normal to fault states. This dataset is widely used in aero-engine fault prediction and remaining service life analysis research. This dataset consists of multiple multivariate time-series signals, including sampling data from 21 sensors, namely: total fan inlet temperature, total low-pressure compressor outlet temperature, total high-pressure compressor outlet temperature, total low-pressure turbine outlet temperature, fan inlet pressure, and total bypass duct pressure.
[0020] Figure 1 In the middle, there is a fan, an LPC low-pressure compressor, an LPT low-pressure turbine, an HPC high-pressure compressor, an HPT high-pressure turbine; a combustor, a nozzle; N1 low-pressure rotor speed, and N2 high-pressure rotor speed.
[0021] The technical solution of this application is as follows: Please see Figure 2 This application provides a method for predicting aircraft engine failures, including: S01. Obtain the characteristic variables; S02. Based on the characteristic variables, use the MAF method to perform minimum / maximum autocorrelation factor analysis and extract strong autocorrelation factors; S03. Construct a training set using the extracted strong autocorrelation factors and their corresponding remaining lifetimes; S04. Obtain a support vector regression model using the training set; S05. Use support vector regression models to predict the remaining service life of aero engines online.
[0022] In S01: In some embodiments, characteristic variables are obtained through wrap-around variable selection, and the characteristic variables include the total outlet temperature of the high-pressure compressor, denoted as... The total outlet temperature of the low-pressure turbine is set to... The fan inlet pressure is set to... .
[0023] It is understandable that there are 21 measurement variables used to measure the state of an aero-engine, but some variables are significantly affected by noise, while others are irrelevant to the remaining service life. Using measurement data from all variables to construct a degradation model might lead to a deviation from reality. This application utilizes wrapper variable selection to select variables. Wrapper variable selection is similar to cross-validation algorithms; its core idea is to use the final model performance as an evaluation criterion for a subset of features, and to iteratively optimize the selection of features.
[0024] In S02: In some embodiments, the method for extracting strong autocorrelation factors includes the following steps: S021. Let the original space variable be... In the formula, This refers to the total outlet temperature of the high-pressure compressor. This refers to the total outlet temperature of the low-pressure turbine. Let t be the fan inlet pressure and t be the current time; let the time lag parameter be... For the first A sample matrix consisting of time series The sample matrix consists of rows representing samples and columns representing state variables. A time lag parameter is defined. ; S022, For Find the state difference matrix of the aero-engine. The formula is as follows: ; In the formula, t represents the current time. For time lag parameters, For the first A time series, For the first A sample matrix consisting of time series; S023, Calculate the variance-covariance matrix and ,as follows: ; ; In the formula, t represents the current time. For time lag parameters, For the first A time series, For the first A sample matrix consisting of n time series, where n is the number of samples in the sample matrix, and T is the transpose of the matrix.
[0025] S024, Calculate the matrix * ,in, express The false reversal; S025. Solve the following eigenvalue decomposition problem: ( * ) = ; In the formula, The covariance matrix is the sample lag time with zero. Lag time The covariance matrix, for The false rebellion, For the first A time series, Let λ be the eigenvector and λ be the eigenvalue.
[0026] In S025: In some embodiments, solving the following eigenvalue decomposition problem includes: S0251, The eigenvalues are arranged in ascending order, that is: The eigenvectors corresponding to these eigenvalues are the projection vectors of the autocorrelation factors, respectively. Let the projection vector be... When, i.e., the projection vector is For the original space sample The autocorrelation factor projected onto the projection vector The autocorrelation coefficient of this autocorrelation factor Obviously there is ; S0252. Arrange the correlation coefficients from largest to smallest, and set them as follows: The corresponding projection vectors are respectively Given a threshold ,like Then the former Factors , , ..., It exhibits relatively strong autocorrelation, reflecting the autocorrelation characteristics of the system process state, and is called a strong autocorrelation factor; later Factors with weak autocorrelation reflect noise in the system and are called weak autocorrelation factors.
[0027] It is understandable that step S025 is to explain what a strong autocorrelation factor is.
[0028] Furthermore, strong autocorrelation factors are extracted, and the projection matrices of each strong autocorrelation factor are obtained. ; ; In the formula, 3 represents the number of strongly autocorrelated factors, with 3 representing 3 variables.
[0029] It is understandable that the aircraft engine is in a certain state. , and Forming new samples .
[0030] Furthermore, for a new sample representing the state of an aircraft engine... , Strong autocorrelation factor It can be calculated using the following transformation formula: .
[0031] In S03: Please see Figure 3 In some embodiments, the method for constructing the training set includes: S031, Let the first A data matrix consisting of time series , , The data matrix of the time series, composed of strong autocorrelation components, was obtained after analysis using the MAF method. , , The time series is decomposed into non-overlapping windows of size L, where L is a parameter; The Each window is composed of Composition; from each window, extract two parameters for each dimension: the average value within the i-th window. i and the linear regression trend coefficient on the window i Therefore, each window can be calculated to have a size of eigenvectors, where This represents the number of strongly autocorrelated factors in the MAF analysis; S032. Associate the remaining lifetime with each feature vector: For The Each window has a corresponding remaining lifespan calculated as follows: For the entire time series set Repeat the above steps to obtain the training set. ,in This is a vector composed of the features of each hadron correlation factor extracted within each window. It represents the remaining lifetime of the time series starting from the last moment of the corresponding window.
[0032] Strong autocorrelation factors extracted from a training sample composed of historical data. Window division (parameters) and feature extraction such as Figure 3 As shown. From Figure 3 It can be seen that for the average value within the first window 1 and the linear regression trend coefficient on the window 1, 1 and 1 component t1 is the remaining time until the failure time of this window, and the same applies to subsequent windows.
[0033] It is understandable that modeling the degradation process involves linking the state of an aero-engine with its remaining service life. The training dataset (historical data) typically contains the time series of an aero-engine from its normal state to its failure. Due to noise or random errors in the measurements, it is difficult to establish a stable and effective functional relationship between the instantaneous measurements of the state and the remaining service life.
[0034] This application divides the time series data in the training set into windows of a certain size to increase the stability of the degradation model, and extracts the statistical characteristics of the variables in each window. Since the degradation process is dynamic and the variables exhibit autocorrelation, factors with strong autocorrelation are better able to characterize the temporal correlation of the system's state degradation process. This application performs MAF analysis on the original variables to obtain strongly correlated factors. This application proposes window division based on these strongly autocorrelation factors, calculating the mean and slope of each strongly correlated factor within the window. The resulting features describe both the numerical magnitude of the time series data of the strongly autocorrelation factors within a given window and reflect their changing trends.
[0035] In S04: It is understandable that obtaining a support vector regression (SVR) model using a training set is a mature technique in this field, and will not be elaborated upon here.
[0036] In S05: It is understandable that a support vector regression (SVR) model built using historical data can predict the remaining service life of an aircraft engine; monitoring the operating status of an online aircraft engine aims to predict the time when the component may fail based on past experience.
[0037] In some embodiments, the online prediction of the remaining service life of an aircraft engine using a support vector regression (SVR) model includes the following steps: S051. Assume the aircraft engine testing process is a time series. Let the current time be the last monitoring time. ; S052, Time Series After variable screening, MAF analysis generates time series containing strongly autocorrelated variables. For the current time point, we take... - L to The strongly autocorrelated variable sequences are used as the time window, where the window length is set to L=20, and the window feature vector is extracted and set as follows. ,Will Applied to the Support Vector Regression (SVR) model, it obtains the remaining service life of the aircraft engine under the current monitoring status, i.e.: ,in This is a support vector regression model.
[0038] It is understandable that we assume the current time is the last monitoring time. If no relevant information is available at this point, the component may be in the early, late, or any stage in between of its lifecycle. Remaining useful life (RUL) estimation aims to predict the last monitoring point. The time difference between the moment the test equipment malfunctions and the moment it fails.
[0039] Secondly, embodiments of this application provide an aircraft engine fault prediction system, comprising: The feature variable module is used to obtain feature variables; The strong autocorrelation factor module uses the MAF method to perform minimum / maximum autocorrelation factor analysis based on characteristic variables to extract strong autocorrelation factors. The training set construction module utilizes the extracted strong autocorrelation factors and their corresponding remaining lifetimes to construct the training set; The Support Vector Regression Model module uses the training set to obtain the support vector regression model. The online prediction module uses a support vector regression model to predict the remaining service life of aero engines online.
[0040] In the feature variable module: In some embodiments, characteristic variables are obtained through wrap-around variable selection, and the characteristic variables include the total outlet temperature of the high-pressure compressor, denoted as... The total outlet temperature of the low-pressure turbine is set to... The fan inlet pressure is set to... .
[0041] It is understandable that there are 21 measurement variables used to measure the state of an aero-engine, but some variables are significantly affected by noise, while others are irrelevant to the remaining service life. Using measurement data from all variables to construct a degradation model might lead to a deviation from reality. This application utilizes wrapper variable selection to select variables. Wrapper variable selection is similar to cross-validation algorithms; its core idea is to use the final model performance as an evaluation criterion for a subset of features, and to iteratively optimize the selection of features.
[0042] In the strongly autocorrelation factor module: In some embodiments, the method for extracting strong autocorrelation factors includes the following steps: S021. Let the original space variable be... In the formula, This refers to the total outlet temperature of the high-pressure compressor. This refers to the total outlet temperature of the low-pressure turbine. Let t be the fan inlet pressure and t be the current time; let the time lag parameter be... For the first A sample matrix consisting of time series The sample matrix consists of rows representing samples and columns representing state variables. A time lag parameter is defined. ; S022, For Find the state difference matrix of the aero-engine. The formula is as follows: ; In the formula, t represents the current time. For time lag parameters, For the first A time series, For the first A sample matrix consisting of time series; S023, Calculate the variance-covariance matrix and ,as follows: ; ; In the formula, t represents the current time. For time lag parameters, For the first A time series, For the first A sample matrix consisting of n time series, where n is the number of samples in the sample matrix, and T is the transpose of the matrix.
[0043] S024, Calculate the matrix * ,in, express The false reversal; S025. Solve the following eigenvalue decomposition problem: ( * ) = ; In the formula, The covariance matrix is the sample lag time with zero. Lag time The covariance matrix, for The false rebellion, For the first A time series, Let λ be the eigenvector and λ be the eigenvalue.
[0044] In S025: In some embodiments, solving the following eigenvalue decomposition problem includes: S0251, The eigenvalues are arranged in ascending order, that is: The eigenvectors corresponding to these eigenvalues are the projection vectors of the autocorrelation factors, respectively. Let the projection vector be... When, i.e., the projection vector is For the original space sample The autocorrelation factor projected onto the projection vector The autocorrelation coefficient of this autocorrelation factor Obviously there is ; S0252. Arrange the correlation coefficients from largest to smallest, and set them as follows: The corresponding projection vectors are respectively Given a threshold ,like Then the former Factors , , ..., It exhibits relatively strong autocorrelation, reflecting the autocorrelation characteristics of the system process state, and is called a strong autocorrelation factor; later Factors with weak autocorrelation reflect noise in the system and are called weak autocorrelation factors.
[0045] Furthermore, strong autocorrelation factors are extracted, and the projection matrices of each strong autocorrelation factor are obtained. ; ; Furthermore, for a new sample representing the state of an aircraft engine... , Strong autocorrelation factor It can be calculated using the following transformation formula: .
[0046] In the module for constructing the training set: In some embodiments, the method for constructing a training set includes: S031, Let the first A data matrix consisting of time series , , The data matrix of the time series, composed of strong autocorrelation components, was obtained after analysis using the MAF method. , , The time series is decomposed into non-overlapping windows of size L, where L is a parameter; The Each window is composed of Composition; from each window, extract two parameters for each dimension: the average value within the i-th window. i and the linear regression trend coefficient on the window i Therefore, each window can be calculated to have a size of eigenvectors, where This represents the number of strongly autocorrelated factors in the MAF analysis; S032. Associate the remaining lifetime with each feature vector: For The Each window has a corresponding remaining lifespan calculated as follows: For the entire time series set Repeat the above steps to obtain the training set. ,in This is a vector composed of the features of each hadron correlation factor extracted within each window. It represents the remaining lifetime of the time series starting from the last moment of the corresponding window.
[0047] Please see Figure 3 Strong autocorrelation factors extracted from a training sample composed of historical data. Window division (parameters) ) and feature extraction steps as follows Figure 3 As shown.
[0048] It is understandable that modeling the degradation process involves linking the state of an aero-engine with its remaining service life. The training dataset (historical data) typically contains the time series of an aero-engine from its normal state to its failure. Due to noise or random errors in the measurements, it is difficult to establish a stable and effective functional relationship between the instantaneous measurements of the state and the remaining service life.
[0049] This application divides the time series data in the training set into windows of a certain size to increase the stability of the degradation model, and extracts the statistical characteristics of the variables in each window. Since the degradation process is dynamic and the variables exhibit autocorrelation, factors with strong autocorrelation are better able to characterize the temporal correlation of the system's state degradation process. This application performs MAF analysis on the original variables to obtain strongly correlated factors. This application proposes window division based on these strongly autocorrelation factors, calculating the mean and slope of each strongly correlated factor within the window. The resulting features describe both the numerical magnitude of the time series data of the strongly autocorrelation factors within a given window and reflect their changing trends.
[0050] In the support vector regression model module: It is understandable that obtaining a support vector regression (SVR) model using a training set is a mature technique in this field, and will not be elaborated upon here.
[0051] In the online prediction module: It is understandable that a support vector regression (SVR) model built using historical data can predict the remaining service life of an aircraft engine; monitoring the operating status of an online aircraft engine aims to predict the time when the component may fail based on past experience.
[0052] In some embodiments, the online prediction of the remaining service life of an aircraft engine using a support vector regression (SVR) model includes the following steps: S051. Assume the aircraft engine testing process is a time series. Let the current time be the last monitoring time. ; S052, Time Series After variable screening, MAF analysis generates time series containing strongly autocorrelated variables. For the current time point, we take... - L to The strongly autocorrelated variable sequences are used as the time window, where the window length is set to L=20, and the window feature vector is extracted and set as follows. ,Will Applied to the Support Vector Regression (SVR) model, it obtains the remaining service life of the aircraft engine under the current monitoring status, i.e.: ,in This is a support vector regression model.
[0053] It is understandable that we assume the current time is the last monitoring time. If no relevant information is available at this point, the component may be in the early, late, or any stage in between of its lifecycle. Remaining useful life (RUL) estimation aims to predict the last monitoring point. The time difference between the moment the test equipment malfunctions and the moment it fails.
[0054] Thirdly, this application provides a computer device including a storage device and a processor, wherein the storage device stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the aircraft engine fault prediction method described above.
[0055] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.
[0056] The memory includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or D-interface display memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory may be an internal storage unit of the computer device, such as the hard disk or memory of the computer device. In other embodiments, the memory may also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the computer device. Of course, the memory may include both the internal storage unit and the external storage device of the computer device. In this embodiment, the memory is often used to store the operating system and various application software installed on the computer device, such as the program code of the aircraft engine fault prediction method. In addition, the memory can also be used to temporarily store various types of data that have been output or will be output.
[0057] In some embodiments, the processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor is typically used to control the overall operation of the computer device. In this embodiment, the processor is used to run program code stored in the memory or process data, for example, to run the program code of the aircraft engine fault prediction method.
[0058] Fourthly, this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the aircraft engine fault prediction method described above.
[0059] The computer-readable storage medium stores an interface display program that can be executed by at least one processor to cause the at least one processor to perform the steps of the aircraft engine fault prediction method described above.
[0060] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the aircraft engine fault prediction method described in the embodiments of this application.
[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for predicting aircraft engine failures, characterized in that, include: Obtain the characteristic variables; Based on the characteristic variables, the MAF method was used to perform minimum / maximum autocorrelation factor analysis to extract strong autocorrelation factors. A training set was constructed using the extracted strong autocorrelation factors and their corresponding remaining lifetimes; Use the training set to obtain a support vector regression model; The remaining service life of aero engines is predicted online using a support vector regression model.
2. The aircraft engine fault prediction method according to claim 1, characterized in that, Feature variables are obtained through wrap-around variable selection. These feature variables include the total outlet temperature of the high-pressure compressor, denoted as [missing value]. The total outlet temperature of the low-pressure turbine is set to... The fan inlet pressure is set to... .
3. The aircraft engine fault prediction method according to claim 1, characterized in that, The method for extracting strong autocorrelation factors includes the following steps: Let the original space variable be In the formula, This refers to the total outlet temperature of the high-pressure compressor. This refers to the total outlet temperature of the low-pressure turbine. Let t be the fan inlet pressure and t be the current time; let the time lag parameter be... For the first A sample matrix consisting of time series The sample matrix consists of rows representing samples and columns representing state variables. A time lag parameter is defined. ; for Find the state difference matrix of the aero-engine. The formula is as follows: ； In the formula, t represents the current time. For time lag parameters, For the first A time series, For the first A sample matrix consisting of time series; Calculate the variance-covariance matrix and ,as follows: ； ； In the formula, t represents the current time. For time lag parameters, For the first A time series, For the first A sample matrix consisting of n time series, where n is the number of samples in the sample matrix and T is the transpose of the matrix; Calculate matrix * ,in, express The false reversal; Solve the following eigenvalue decomposition problem: （ * ） = ； In the formula, The covariance matrix is the sample lag time with zero. Lag time The covariance matrix, for The false rebellion, For the first A time series, Let λ be the eigenvector and λ be the eigenvalue.
4. The aircraft engine fault prediction method according to claim 3, characterized in that, Solve the following eigenvalue decomposition problems, including: The eigenvalues are arranged in ascending order, that is: The eigenvectors corresponding to these eigenvalues are the projection vectors of the autocorrelation factors, respectively. Let the projection vector be... When, i.e., the projection vector is For the original space sample The autocorrelation factor projected onto the projection vector The autocorrelation coefficient of this autocorrelation factor Obviously there is ; Arrange their respective correlation coefficients from largest to smallest, and set them as follows: The corresponding projection vectors are respectively Given a threshold ,like Then the former Factors , , ..., It exhibits relatively strong autocorrelation, reflecting the autocorrelation characteristics of the system process state, and is called a strong autocorrelation factor; later Factors with weak autocorrelation reflect noise in the system and are called weak autocorrelation factors.
5. The aircraft engine fault prediction method according to claim 4, characterized in that, Extract the strong autocorrelation factors and obtain the projection matrix of each strong autocorrelation factor. ； ； For a new sample representing the state of an aircraft engine , Strong autocorrelation factor It can be calculated using the following transformation formula: 。 6. The aircraft engine fault prediction method according to claim 1, characterized in that, Methods for constructing training sets include: Let the first A data matrix consisting of time series , , The data matrix of the time series, composed of strong autocorrelation components, was obtained after analysis using the MAF method. , , The time series is decomposed into non-overlapping windows of size L, where L is a parameter; The Each window is composed of Composition; from each window, extract two parameters for each dimension: the average value within the i-th window. i and the linear regression trend coefficient on the window i Therefore, each window can be calculated to have a size of eigenvectors, where This represents the number of strongly autocorrelated factors in the MAF analysis; Associate the remaining lifetime time with each feature vector: for The Each window has a corresponding remaining lifespan calculated as follows: For the entire time series set Repeat the above steps to obtain the training set. ,in This is a vector composed of the features of each hadron correlation factor extracted within each window. It represents the remaining lifetime of the time series starting from the last moment of the corresponding window.
7. The aircraft engine fault prediction method according to claim 1, characterized in that, The online prediction of the remaining service life of aero-engines using a support vector regression model includes the following steps: Assume the aircraft engine testing process is a time series. Let the current time be the last monitoring time. ; Time series After variable screening, MAF analysis generates time series containing strongly autocorrelated variables. For the current time point, we take... - L to The strongly autocorrelated variable sequences are used as the time window, where the window length is set to L=20, and the window feature vector is extracted and set as... ,Will Applied to the Support Vector Regression (SVR) model, it obtains the remaining service life of the aircraft engine under the current monitoring status, i.e.: ,in This is a support vector regression model.
8. An aircraft engine fault prediction system, characterized in that, include: The feature variable module is used to obtain feature variables; The strong autocorrelation factor module uses the MAF method to perform minimum / maximum autocorrelation factor analysis based on characteristic variables to extract strong autocorrelation factors. The training set construction module utilizes the extracted strong autocorrelation factors and their corresponding remaining lifetimes to construct the training set; The Support Vector Regression Model module uses the training set to obtain the support vector regression model. The online prediction module uses a support vector regression model to predict the remaining service life of aero engines online.
9. A computer device, characterized in that, It includes a storage device and a processor, the storage device storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The device stores a computer program that, when executed by a processor, causes the processor to perform the steps of the method as described in any one of claims 1-7.