Data-driven method for fault monitoring and identification of the entire aircraft engine system

Abstract

The present invention provides a data-driven method for monitoring and identifying faults in the entire system of an aero-engine, including: collecting multivariate time series data of the engine under various flight conditions; establishing a system of performance parameters for the entire system of an aero-engine under all operating conditions; establishing a digital twin model of aero-engine performance; reconstructing the performance parameters of the entire system and all operating conditions of the engine, calculating the deviation values ​​of high-dimensional performance parameters, and realizing monitoring and identification of faults in the entire system of an aero-engine. The present invention integrates engine health management with digital twin technology, and utilizes the advantages of deep learning in mining the spatiotemporal correlation of high-dimensional time series data of engines. It proposes a data-driven digital twin model of the performance of the entire system of an engine under all operating conditions, and establishes a hierarchical indicator system for the entire system of the entire machine, realizing real-time status monitoring and fault identification of the entire flight segment of the engine under dynamic operating conditions and environmental conditions, and providing model and algorithm support for the health management of aero-engines based on digital twin technology.

Description

Technical Field

[0001] The present invention relates to the field of aircraft engine health management, and in particular to a data-driven aircraft engine full-system fault monitoring and identification method. Background Art

[0002] Aircraft engine failures can cause significant economic damage and potentially high social costs. Engine condition monitoring systems are used to detect, isolate, and identify long-term trends in performance degradation and faults. Therefore, engine health monitoring and abnormality monitoring are critical, which places higher demands on engine performance condition monitoring.

[0003] In modern aircraft, full flight data continuously sampled throughout the flight provides an opportunity to achieve a wider range of fault detection. In practical applications, the entire flight data is rarely analyzed. As a minimum requirement for aircraft engine health monitoring, most airlines use discrete data points collected during takeoff and cruise for engine condition monitoring. However, these discrete data are sparse, which makes it difficult to distinguish randomly scattered single faults, which may lead to delays in detection performance degradation. Depending on the faulty component and the severity of the fault, multiple flights may be required to obtain multiple data points to detect the fault, which may cause alarms and false alarms in the process. Therefore, the entire flight segment should be monitored continuously to improve the fault detection rate. Currently, most airlines use quick access recorders (QARs) for data acquisition, providing flight data continuously sampled at 1Hz or higher, also known as full flight data. The availability of these data sets enables the introduction of new methods for engine condition monitoring, which provides an opportunity to detect engine faults in a single flight.

[0004] At present, most of them adopt the method of model-driven and data-driven fusion. Data and models are the core elements of building digital twins of aircraft engines. The digital twin system architecture of aircraft engines is constructed. This architecture models the engine body and the engine operating environment separately, while the digital twin system combining physical models and data-driven often requires the support of professional knowledge. An alternative method for health monitoring is to use data-driven models. With the continuous development of digital control systems and sensor measurement technologies for aircraft engines, engine status monitoring data has grown exponentially and has the typical characteristics of high speed, multi-source, and heterogeneity. Machine learning, especially deep learning, has great potential in fault detection of time series data, and has the ability to learn fault modes directly from raw sensor data and handle nonlinearity in complex time correlations. A digital twin model of aircraft engine performance based on deep learning is constructed. Of course, compared with traditional data-driven methods, data-driven methods based on deep learning pay more attention to system knowledge and can distinguish the impact of changes in operating conditions and the impact of degradation on sensor readings in the case of limited labeled data.

[0005] Currently, the most commonly used method for aircraft engine health monitoring is based on gas path analysis (GPA). Existing methods can be divided into two categories: physics-based methods and data-driven methods. Physics-based methods aim to describe the physics of failure mechanisms by mathematically modeling the components and systems under study. Data-driven methods mainly rely on key performance parameters of the gas path subsystem. For example, based on takeoff and cruise message data, mathematical statistics, machine learning, and deep learning models are used to analyze key performance parameters such as exhaust temperature deviation, maximum exhaust temperature, fuel flow deviation, and high-pressure rotor speed deviation, so as to perform simple monitoring of the health status of aircraft engines.

[0006] In summary, the existing technology lacks an engine performance status monitoring method that can fully tap the value of engine monitoring data and provide support for engine health monitoring and abnormality detection. Summary of the invention

[0007] Purpose of the invention: The technical problem to be solved by the present invention is to provide a data-driven method for monitoring and identifying faults in the entire system of an aircraft engine, in order to achieve health monitoring and abnormality detection of the aircraft engine. The method comprises:

[0008] Step 1: dividing the flight data according to the flight conditions, and collecting the time series data of the engine performance parameters corresponding to each flight condition in the current flight cycle;

[0009] Step 2: According to the system composition principle of aircraft engines, extract the time series data of performance parameters of subsystems (generally four main subsystems (gas path, lubricating oil, fuel and air)) to establish a performance parameter system for the entire system and all operating conditions of aircraft engines.

[0010] Step 3: The multivariate time series data of the engine has temporal and spatial correlation, and has the natural attribute of "temporal coherence" on the time axis. From a spatial perspective, due to the strong coupling between engine performance parameters and environmental parameters, it has a strong "spatial correlation". The present invention utilizes the advantages of deep learning models in high-dimensional time series data analysis, explores the temporal and spatial correlation of high-dimensional time series data of the entire aero-engine system, and establishes a data-driven aero-engine performance digital twin model covering the entire aero-engine system and all operating conditions.

[0011] Step 4: Use the digital twin model of aircraft engine performance to reconstruct the key performance parameters of the subsystem under all operating conditions, and calculate the reconstruction deviation value sequence of the performance parameters under all operating conditions and all flight segments.

[0012] Step 5. Establish a hierarchical indicator system for whole-machine health monitoring based on the digital twin of engine performance. On the basis of reconstructing the deviation value sequence of performance parameters under all working conditions and all flight segments, use the time series feature extraction method to extract features of the deviation value sequence of key performance parameters, determine the state indicators of the parameters, and then obtain the subsystem state indicators. Finally, obtain the indicator that characterizes the overall health status of the entire aircraft engine system, namely, the whole-machine digital twin health index HI.

[0013] In step 1, the engine performance parameter time series data includes low-pressure rotor speed, high-pressure rotor speed, lubricating oil temperature, lubricating oil pressure, lubricating oil flow, exhaust temperature, fuel metering valve, fuel flow, throttle lever thrust gradient, variable bypass valve position, high-pressure compressor inlet temperature, high-pressure compressor outlet temperature, variable stator blade position, and operating parameters, and the operating parameters include static temperature, airspeed, Mach number and altitude.

[0014] Furthermore, in step 1:

[0015] According to the characteristics of the monitored object, the performance parameters under appropriate working conditions are selected for the training of the system reconstruction model. In a complete flight cycle, the working state of the engine under different flight conditions is different, and the monitored performance parameters fluctuate greatly. The selection of performance parameters under different working conditions has an important impact on the health management of aircraft engines. Considering that the engine performance parameters will show a strong performance correlation under high-power flight conditions, the flight conditions include take-off conditions, initial climb conditions, climb conditions and cruise conditions. ICAO and the Commercial Aviation Safety Group have issued a guidance document on the division of aircraft flight phases (Phase of Flight: Definitions and Usage Notes). The guidelines clearly define the characteristics and subordinate sub-phases of daily flight phases of civil aircraft such as ground docking, push-out / tow, taxiing, take-off, initial climb, cruise, approach and landing; the judgment logic of the flight phase is determined according to the state transition of the aircraft. The judgment logic will detect the current flight phase state of the aircraft and determine whether the aircraft enters the next flight stage through specific parameter judgment conditions, such as flight altitude parameters. These flight phases are in a high-power state, and the engine performance parameters will show a strong performance correlation during the high-power phase.

[0016] Furthermore, step 2 includes:

[0017] The subsystems include an engine air path subsystem, a lubricating oil subsystem, a fuel subsystem and an air subsystem, and the performance parameter time series data is used for training a performance parameter reconstruction model.

[0018] The digital twin model of the engine performance is established based on the local linear model of the main subsystems of the aircraft engine. According to the system composition principle of the aircraft engine, the performance parameters of the four main subsystems of the engine gas path subsystem, lubricating oil system, fuel system and air system are extracted to establish the performance parameter system of the entire system and all working conditions of the aircraft engine, providing a data basis for the construction of the digital twin model of the aircraft engine performance.

[0019] Furthermore, the performance parameters and environmental parameter data include: engine high-pressure rotor speed, engine low-pressure rotor speed, high-pressure compressor inlet temperature, high-pressure compressor outlet temperature, PS3 selection, engine exhaust temperature, throttle lever thrust gradient, fuel metering valve, fuel flow, lubricating oil temperature, lubricating oil pressure, lubricating oil flow, variable stator vane position, variable bypass valve position, flight altitude, flight Mach number, total atmospheric temperature, static temperature, airspeed, flight altitude, engine anti-icing status, and wing anti-icing status.

[0020] Furthermore, step 3 includes:

[0021] Engine data usually covers system status, operating environment and working condition data. Due to the strong coupling between the aircraft system status and flight conditions, in a dynamic and changeable operating environment, engine data is usually high-dimensional time series data with strong dynamic spatiotemporal correlation. Therefore, we use the advantages of deep learning models in processing high-dimensional spatiotemporal information (time series) data to explore the implicit feature correlation and inherent complexity law (spatiotemporal correlation) of the high-dimensional spatiotemporal sequence data of the entire aircraft engine system. Such as LSTM-AE, LSTM-VAE and other models.

[0022] Furthermore, the engine time series data monitored under healthy conditions are first collected, and the input data is standardized and reorganized into an M*N matrix structure, where M is the set sliding window size, and N is the number of characteristic parameters related to the performance parameters. For example, the parameter input matrix x at time t with a sliding window of 20 is:

[0023]

[0024] Where Nn(t+10) represents the data matrix of the nth parameter related to the performance parameter at the 10th second after time t;

[0025] The matrix x contains the data at the current time t, a total of 10 seconds before and after, and a total length of 21 seconds. This is the concept of a sliding window. The matrix here can be regarded as a "window" moving on the data sequence, and the elements in the window represent the state of the aircraft engine within a certain time interval.

[0026] Then, the powerful function of LSTM (Long Short-Term Memory Neural Network) in processing time series data and the learning ability of AE in dimensionality reduction and feature extraction are effectively utilized. The parameter input matrix x maps the high-dimensional input x to the low-dimensional hidden state h under the action of the encoder: h = f(W·x+b), thereby forcing the neural network to learn the most informative features, where the function f includes the weight matrix W representing the input layer and the hidden layer and the encoded bias b; the autoencoder network uses h to reconstruct the original data to obtain the reconstructed data x*: x * =g(W′·h+b′), where the function g includes the weight matrix W′ between the hidden layer and the output layer and the decoding bias b′;

[0027] The encoder and decoder are generally parameterized equations. Finally, a loss error function δ = [W, b, W′, b′] is set to minimize the error and achieve model optimization. Under the LSTM-AE framework, the construction of a data-driven digital twin model of aviation engine performance covering the entire system and all operating conditions of the aviation engine is completed.

[0028] Furthermore, step 4 includes: using the trained and optimized LSTM-AE digital twin model to reconstruct the performance parameters of the four main subsystems under all working conditions, and calculating the reconstruction deviation value sequence Y using the following formula:

[0029]

[0030] in represents the predicted value of the performance parameters by the performance digital twin model at time t, Y t is the measured value of the parameter.

[0031] In step 5, based on the deviation value sequence of the key performance parameters of the whole system and all working conditions obtained by the performance digital twin, the time series feature extraction method is used to extract the characteristics of the deviation value sequence, such as mean square error, relative average error, quantile, fitting distribution, variance, slope and maximum value, that is, the state index of the performance parameter, and standardize it, 0 means intact, 1 means scrapped, and the connection between the subsystem and the whole system of the aircraft engine is used to sum to obtain the health index HI of the whole digital twin. The larger the health index HI of the whole digital twin, the more seriously degraded the engine is. A multi-level health monitoring indicator system covering the whole system and all working conditions of the aircraft engine is established to realize real-time monitoring and precise positioning of all working condition faults during the flight process.

[0032] In step 5, the time series feature extraction method is used to extract the features of the deviation value sequence, including the root mean square error (RMSE) and relative error (RE):

[0033]

[0034] Where T represents the total time of the test flight.

[0035] In step 5, for a parameter of the gas path subsystem, the root mean square error (RMSE) is used as the characteristic index, and the state index of the parameter is R 1i , the gas circuit subsystem status index is R1, and the sum is used to obtain the health index HI of the whole machine digital twin:

[0036] R 1i =RMSE(XX*)

[0037]

[0038] Where X represents the input data, X * represents the reconstructed output data, P is the number of all key performance parameters of the gas path subsystem, Q represents the number of subsystems, R 1i represents the i-th state index of the gas circuit subsystem, R j Represents the status indicator of the jth subsystem in the entire aircraft engine system.

[0039] The beneficial effects of the present invention are:

[0040] 1. Establish a performance parameter system for the entire aircraft engine system and all operating conditions:

[0041] In order to solve the problem of large volatility in aircraft engine monitoring data throughout the flight cycle, the present invention selects high-power data consisting of take-off conditions, initial climb conditions, climb conditions and cruise conditions; at the same time, according to the system composition principle of the aircraft engine, performance parameters of four main subsystems (engine air path subsystem, lubricating oil system, fuel system and air system) are extracted; combined with the environmental data in the engine operating environment, the three types of data build a typical aircraft engine full-system and full-condition performance parameter system.

[0042] 2. Establish a data-driven digital twin model of aircraft engine performance:

[0043] Engine data comes from a wide range of sources and has the characteristics of multi-source heterogeneity, multi-modality, and multiple spatiotemporal scales. Multiple spatiotemporal scales can also be understood as spatiotemporal correlation. The engine data has "temporal coherence" on the time axis, and from a spatial perspective, due to the strong coupling between engine performance parameters and environmental parameters, it has a strong "spatial correlation". The present invention deeply integrates the multi-layer automatic feature extraction expression capability of the deep learning model with the nonlinear, high-dimensional and spatiotemporal correlation analysis methods of the engine spatiotemporal series data to establish a digital twin model of aero-engine performance based on a deep learning model.

[0044] 3. Establish a hierarchical monitoring indicator system for the health monitoring of the entire aircraft engine system:

[0045] The present invention utilizes a time series feature extraction method to perform feature extraction on a sequence of deviation values ​​of key performance parameters, determines the status indicators of the performance parameters, and then obtains the subsystem status indicators, and finally obtains an indicator characterizing the overall health status of the entire aero-engine system, namely, the digital twin health index HI of the entire machine. A hierarchical monitoring indicator system is established from the basic performance parameter status indicators to the digital twin health indicators of the entire machine, to achieve real-time monitoring and precise positioning of all operating condition faults during the flight process.

[0046] In summary, the present invention focuses on the integration of engine health management and digital twin technology, and utilizes the advantages of deep learning in mining the spatiotemporal correlation of high-dimensional time series data of the engine. It proposes a data-driven digital twin model of the engine's full system and full operating condition performance, and establishes a hierarchical indicator system for the entire aircraft engine system to achieve real-time status monitoring and fault identification of the engine under dynamic operating conditions and environmental conditions throughout the entire flight segment, providing model and algorithm support for aircraft engine health management based on digital twin technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0048] Figure 1 Flow chart of the method of the present invention.

[0049] Figure 2 Schematic diagram of aircraft engine subsystems and key performance parameters in an embodiment of the present invention.

[0050] Figure 3 This is a schematic diagram of the structure of the engine performance digital twin model based on LSTM-AE in an embodiment of the present invention.

[0051] Figure 4 This is the process of establishing a performance model using a sliding window for EGT parameters in an embodiment of the present invention.

[0052] Figure 5 This is a comparison diagram of parameter reconstruction in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The embodiment of the present invention provides a data-driven digital twin modeling method for engine multi-operating condition performance, the flow chart is as follows: Figure 1 As shown, including:

[0054] Step 1: The flight data is divided according to the flight conditions, and the corresponding multivariate time series data of the engine in each flight condition of the current flight cycle are collected. In a complete flight cycle, the working state of the engine is different under different flight conditions, and the monitored performance parameters fluctuate greatly. Considering that the engine performance parameters will show a strong performance correlation in the high-power stage, the data of the take-off condition, the initial climb condition, the climb condition and the cruise condition are selected in the present invention for the study of the performance of the aircraft engine, so as to achieve the coverage of the working state of the engine in all conditions and all segments.

[0055] Step 2: According to the system composition principle of the aircraft engine, the parameters of the four main subsystems (gas path, lubricating oil, fuel and air) are extracted to establish the performance parameter system of the aircraft engine in the whole system and all working conditions, such as Figure 2 As shown, the sensor locations of some performance parameters are marked.

[0056] These parameters include key performance parameters and environmental parameters of each system. Taking the exhaust temperature parameter EGT in the gas path subsystem as an example, through correlation analysis and system principle knowledge, the parameters related to this performance parameter include: engine high-pressure rotor speed, engine low-pressure rotor speed, PS3 selection, high-pressure compressor inlet temperature, high-pressure compressor outlet temperature of the gas path subsystem; lubricating oil temperature, lubricating oil pressure, lubricating oil flow of the lubricating oil system; throttle lever thrust gradient, fuel metering valve, fuel flow, fuel-air ratio saturation of the fuel system, and variable stator blade position and variable bypass valve position of the air system. It also includes environmental parameters such as static temperature, flight airspeed, Mach number and flight altitude.

[0057] Step 3: Engine data is usually high-dimensional time series data with strong dynamic spatiotemporal correlation. By leveraging the advantages of deep learning models in high-dimensional time series data analysis, we can explore the implicit feature correlation and inherent complexity laws (spatiotemporal correlation) of the high-dimensional spatiotemporal sequence data of the entire aircraft engine system, and establish a data-driven aircraft engine performance digital twin model covering the entire aircraft engine system and all operating conditions.

[0058] For example, using the LSTM (Long Short-Term Memory) model to process long time series and the advantages of the AE model in unsupervised feature learning, Figure 3 The LSTM-AE-based digital twin model of aviation engine performance is shown.

[0059] First, the engine time series data monitored under healthy conditions are collected, and the input data is standardized and reorganized into an M*N matrix structure, where M is the set sliding window size and N is the number of characteristic parameters related to the performance parameters. For example, the input matrix of a parameter at time t with a sliding window of 20 is:

[0060]

[0061] The matrix X contains the data at the current time t, a total of 10 seconds before and after, and a total length of 21 seconds. This is the concept of a sliding window. The matrix here can be regarded as a "window" moving on the data sequence, and the elements in the window represent the state of the aircraft engine within a certain time interval.

[0062] Then, the powerful function of LSTM in processing time series data and the learning ability of AE in dimensionality reduction and feature extraction are effectively utilized. The parameter input matrix x maps the high-dimensional input x to the low-dimensional hidden state h under the action of the encoder, that is, h = f(W·x+b), thereby forcing the neural network to learn the most informative features, where the function f includes the weight matrix W representing the input layer and the hidden layer and the encoded bias b; the autoencoder network uses h to reconstruct the original data to obtain the reconstructed data x*: x * =g(W′·h+b′), where the function g includes the weight matrix W′ between the hidden layer and the output layer and the decoding bias b′. Figure 3 shown.

[0063] The encoder and decoder are generally parameterized equations. Finally, a loss error function δ = [W, b, W′, b′] will be set to minimize the error and achieve model optimization. Then, under the LSTM-AE framework, the digital twin model of the full system and full operating condition performance of the aircraft engine based on the deep learning model is completed.

[0064] Taking EGT performance parameters as an example, the process of establishing its performance model is as follows: Figure 4 As shown, the processed EGT time series data is input, and the input data is converted into a time series matrix. The sliding window value is set to 20, then the length of the time series matrix is ​​20, and the dimension is all parameters related to EGT (see step 2). The matrix contains all relevant parameters except the current t seconds and the previous 19 seconds EGT. In order for the engine performance digital twin model to be applied after the flight, the data after the current time t may provide some useful trend information. Therefore, the present invention designs another matrix form, and this matrix Xt contains the data at the current time t, with a total length of 21 seconds. Then, the data samples obtained under the fault-free state are used for training, and finally the predicted values ​​of the EGT performance parameters are obtained.

[0065] Step 4. Use the digital twin model of the aero-engine based on the LSTM-AE deep learning model to reconstruct the EGT performance parameters of the gas path subsystem and output a sequence of performance parameter reconstruction deviation values, that is, the difference between the actual measured value and the reference baseline. The training data set is constructed from 300 flights randomly selected from approximately 3,000 flights in two years of operation of a turbofan engine, with a total of more than 100,000 data samples (matrix obtained by the sliding window algorithm). The deep learning model is trained using fault-free data samples, the training loss results of the algorithm are calculated, and then the test samples of 30 flights in the remaining data are used for verification, such as Figure 5 As shown, the predicted value of the performance parameter is obtained, and finally the reconstructed deviation value sequence of the target performance parameter EGT at the current moment is calculated.

[0066] Step 5: Based on the EGT performance parameter deviation value sequence of the entire flight, the time series feature extraction method is used to extract the features of the EGT performance parameter deviation value sequence, determine the parameter status index, and the other performance parameters are similar, and then the gas path subsystem status index is obtained, and finally the indicator that characterizes the overall health status of the entire aircraft engine system is obtained, that is, the whole machine digital twin health index HI. For example, a parameter of the gas path subsystem uses the root mean square error (RMSE) as a characteristic indicator, that is, the state index of the parameter is R 1i , then the state index of the gas circuit subsystem is R1, and the sum is the health index HI of the digital twin of the whole machine performance.

[0067] R 1i =RMSE(XX*)

[0068]

[0069] Where X represents the input data, X* represents the reconstructed output data, P is the number of all key performance parameters of the gas path subsystem, Q represents the number of subsystems, R 1i represents the i-th state index of the gas circuit subsystem, R j Represents the status indicator of the jth subsystem in the entire aircraft engine system.

[0070] In a specific implementation, the present application provides a computer storage medium and a corresponding data processing unit, wherein the computer storage medium can store a computer program, and when the computer program is executed by the data processing unit, the invention content of the data-driven aircraft engine full system fault monitoring and identification method provided by the present invention and some or all of the steps in each embodiment can be executed. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0071] Those skilled in the art can clearly understand that the technical solutions in the embodiments of the present invention can be implemented by means of computer programs and their corresponding general hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention can be essentially or partly contributed to the prior art in the form of a computer program, i.e., a software product, which can be stored in a storage medium and includes several instructions for enabling a device including a data processing unit (which can be a personal computer, a server, a single-chip microcomputer, a MUU or a network device, etc.) to execute the methods described in various embodiments of the present invention or certain parts of the embodiments.

[0072] The present invention provides a method for monitoring and identifying faults in the entire system of an aircraft engine based on data-driven. There are many methods and ways to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be considered as the scope of protection of the present invention. All components not specified in this embodiment can be implemented using existing technologies.

Claims

1. A data-driven method for monitoring and identifying faults in the entire aircraft engine system, characterized in that: The following steps are involved: Step 1: dividing the flight data according to the flight conditions, and collecting the time series data of the engine performance parameters corresponding to each flight condition in the current flight cycle; Step 2: According to the system composition principle of the aircraft engine, extract the time series data of the performance parameters of the subsystems and establish the performance parameter system of the entire system and all working conditions of the aircraft engine; Step 3: Establish a data-driven digital twin model of aircraft engine performance that covers the entire aircraft engine system and all operating conditions; Step 4: Use the digital twin model of aircraft engine performance to reconstruct the key performance parameters of the subsystem under all operating conditions, and calculate the reconstruction deviation value sequence of the performance parameters under all operating conditions and all flight segments; Step 5: Establish a hierarchical indicator system for whole-machine health monitoring based on the digital twin of engine performance. On the basis of the reconstructed deviation value sequence of performance parameters under all working conditions and all flight segments, use the time series feature extraction method to extract the features of the deviation value sequence of key performance parameters, determine the state index of the parameter, and then obtain the subsystem state index. Finally, obtain the indicator that characterizes the overall health status of the entire aircraft engine system, that is, the whole-machine digital twin health index HI; Step 3 includes: collecting engine time series data monitored under healthy conditions, standardizing the input data, and reorganizing it into an M*N matrix structure, where M is the set sliding window size, and N is the number of characteristic parameters related to the performance parameters. The parameter input matrix x at time t with a sliding window of 20 is: Where Nn(t+10) represents the data matrix of the nth parameter related to the performance parameter at the 10th second after time t; The parameter input matrix x maps the high-dimensional input x to a low-dimensional hidden state h under the action of the encoder: h = f(W x + b), where the function f includes the weight matrix W representing the input layer and the hidden layer and the encoded bias b; the autoencoder network uses h to reconstruct the original data to obtain the reconstructed data x*: x * =g(W′·h+b′), where the function g includes the weight matrix W′ between the hidden layer and the output layer and the decoding bias b′; A loss error function δ = [W, b, W′, b′] is set to achieve model optimization. Then, under the LSTM-AE framework, a data-driven digital twin model of aero-engine performance covering the entire system and all operating conditions of the aero-engine is constructed. Step 4 includes calculating the reconstruction deviation value sequence Y using the following formula: in represents the predicted value of the performance parameters by the performance digital twin model at time t, Y t is the measured value of the parameter; In step 5, based on the deviation value sequence of key performance parameters of all operating conditions and all flight segments obtained by the performance digital twin, the characteristics of the deviation value sequence are extracted using the time series feature extraction method, and then standardized, with 0 representing intact and 1 representing scrapped. By using the connection between the subsystem and the entire aircraft engine system, the health index HI of the entire digital twin is obtained by summing up, and a multi-level health monitoring indicator system covering the entire system and all operating conditions of the aircraft engine is established, realizing real-time monitoring and precise positioning of all operating condition faults during the flight process; In step 5, the time series feature extraction method is used to extract the features of the deviation value sequence, including the root mean square error RMSE and the relative error RE: Where T represents the total time of the test flight; In step 5, the root mean square error RMSE is used as the characteristic index for the parameters of the gas path subsystem, and the state index of the parameter is R 1i , the gas circuit subsystem status index is R1, and the sum is used to obtain the health index HI of the whole machine digital twin: R 1i =RMSE(X-X * ) Where X represents the input data, X * represents the reconstructed output data, P is the number of all key performance parameters of the gas path subsystem, Q represents the number of subsystems, R 1i represents the i-th state index of the gas circuit subsystem, R j Represents the status indicator of the jth subsystem in the entire aircraft engine system.

2. The method according to claim 1, characterized in that: In step 1, the engine performance parameter time series data includes low-pressure rotor speed, high-pressure rotor speed, lubricating oil temperature, lubricating oil pressure, lubricating oil flow, exhaust temperature, fuel metering valve, fuel flow, throttle lever thrust gradient, variable bypass valve position, high-pressure compressor inlet temperature, high-pressure compressor outlet temperature, variable stator blade position, and operating parameters, and the operating parameters include static temperature, airspeed, Mach number and altitude.

3. The method according to claim 2, characterized in that In step 1, the flight conditions include take-off conditions, initial climb conditions, climb conditions and cruise conditions.

4. The method according to claim 3, characterized in that In step 2, the subsystems include an engine air path subsystem, a lubricating oil subsystem, a fuel subsystem and an air subsystem, and the performance parameter time series data is used to train a performance parameter reconstruction model, including a low-pressure rotor speed, a high-pressure rotor speed, lubricating oil temperature, lubricating oil pressure, lubricating oil flow, exhaust temperature, fuel metering valve, fuel flow, throttle lever thrust gradient, variable bypass valve position, high-pressure compressor inlet temperature, high-pressure compressor outlet temperature, and variable stator blade position.

Images