Aero-engine health index modeling method fusing linear and nonlinear characteristics

Abstract

The invention provides an aero-engine health index modeling method fusing linear and nonlinear characteristics, and the method specifically comprises the following steps: arranging a multi-source sensor detection system, and collecting the operation state multi-source data of an aero-engine; carrying out normalization processing on the collected data, and carrying out reliability screening; based on the screened residual sensor data, extracting nonlinear degradation features by using VAE and extracting linear degradation features by using SPCA; and designing a dynamic fusion weight according to the obtained nonlinear and linear degradation characteristics, and constructing a composite health index in combination with a weighted moving average optimization algorithm. According to the method, by effectively fusing multi-source sensor data and fully utilizing complementary characteristics of a linear method and a nonlinear method, a more reliable composite health index is constructed, and the degradation state of the aero-engine can be intuitively reflected; and an important reference basis is provided for the evaluation of the running state of the aero-engine, the prediction of the residual life and the scientific formulation of a maintenance strategy.

Description

Technical Field

[0001] The present invention relates to the field of health status assessment of aero-engines, and particularly to a method for modeling aero-engine health indicators that integrates linear and non-linear features. Background Art

[0002] An aero-engine is the core power source of an aircraft. As a key component of the aircraft, its operating condition directly affects the flight performance and safety of the aircraft. With the increasing complexity of aircraft equipment and the diversification of operating environments, it is crucial to evaluate the health status of aero-engines and monitor their degradation trends in a timely manner; accurately predicting the remaining useful life (RUL) of the engine can help airlines effectively arrange maintenance, reduce costs, and improve efficiency and safety. In the past few decades, significant progress has been made in RUL prediction technology, and a variety of methods and techniques have been developed. Among them, the RUL prediction method based on health indicators has become a research hotspot due to its intuitiveness and effectiveness.

[0003] Traditional health status assessment methods mostly rely on observational data of single variables or single sensors, which are difficult to comprehensively capture the complex degradation process of aero-engines. With the rapid development of sensor technology, multi-source sensor monitoring systems can record multi-dimensional heterogeneous data of the engine operating state in real time, providing rich information for health indicator modeling. However, how to effectively integrate multi-source sensor data to construct a comprehensive indicator that can accurately reflect the health status of the equipment remains a difficult point in current research.

[0004] Existing multi-source data processing methods usually include independent component analysis, singular value decomposition, factor analysis, etc. based on linear feature extraction, and kernel principal component analysis, long short-term memory network, convolutional neural network, etc. based on non-linear feature extraction. SPCA captures the dominant direction of data through linear dimensionality reduction and is suitable for processing data with strong linear correlation, but it performs poorly in the face of complex non-linear degradation features. VAE, on the other hand, realizes non-linear feature extraction of complex data by constructing a probability distribution in the latent space and can better capture the non-linear degradation patterns of multi-source sensor data.

[0005] In order to fully utilize the complementary advantages of VAE and SPCA in feature extraction, the present invention proposes a method for modeling aero-engine health indicators that integrates linear and non-linear features. This method can construct a comprehensive health indicator with high discrimination and robustness while maintaining the data dimensionality reduction effect, providing an important reference for the health status assessment and maintenance decision-making of aero-engines. Summary of the Invention

[0006] In view of the problems existing in the above-mentioned prior art, the present invention provides a method for modeling the health index of an aero-engine by integrating linear and non-linear features, so as to solve the technical problems of lack of comprehensiveness and robustness in constructing the health index of an aero-engine by multi-source data fusion in the existing methods.

[0007] To achieve the above technical objectives, the present invention provides the following technical solutions:

[0008] A method for modeling the health index of an aero-engine by integrating linear and non-linear features specifically includes the following steps:

[0009] S1. Arrange a multi-source sensor detection system to collect multi-source data on the operating state of the aero-engine;

[0010] S2. Normalize the collected multi-source data on the operating state, and then perform reliability screening using the Pearson coefficient;

[0011] S3. Perform data fusion on the remaining multi-source data on the operating state after screening through a variational autoencoder (VAE) to obtain non-linear degradation features;

[0012] S4. Perform data fusion on the remaining multi-source data on the operating state after screening through a sparse principal component analysis (SPCA) method to obtain linear degradation features;

[0013] S5. According to the obtained non-linear and linear degradation features, design a dynamic fusion weight, and combine it with a weighted moving average optimization algorithm to construct a composite health index.

[0014] Further, step S1 specifically includes:

[0015] S11. Install and connect multiple sensors for measuring different physical quantities on the aero-engine;

[0016] S12. Configure data transmission mode, sampling frequency, and data storage format parameters;

[0017] S13. After determining that the sensors are in a normal working state, collect multi-source data on the operating state of the aero-engine, and its formula expression is:

[0018]

[0019] where A i is the multi-source data on the operating state of aero-engine i; is the operating state data collected by sensor j on aero-engine i at operation time k.

[0020] Further, step S2 specifically includes:

[0021] S21. Normalize the multi-source data of the operating state of the aero-engine collected to avoid the influence of different sensor dimensions. The normalized multi-source data of the operating state is expressed as:

[0022]

[0023] where, U i is the normalized multi-source data of the operating state of aero-engine i; is the normalized operating state data collected by sensor j on aero-engine i at operating time k;

[0024] S22. Use the Pearson coefficient to perform reliability screening on the normalized multi-source data of the operating state, and eliminate the operating state data collected by sensors with small Pearson coefficients, that is, weak correlations. The screened multi-source data of the operating state is denoted as Y = (y i,1 , y i,2 ,..., y i,j ). The calculation formula of the Pearson coefficient is:

[0025]

[0026] where, is the mean of the sum of historical time values of aero-engine i, is the kth operating time of aero-engine i, is the mean of the normalized operating state data collected by sensor j on engine i; K is the maximum operating time of aero-engine i.

[0027] Further, step S3 specifically includes:

[0028] S31. Use the screened multi-source data of the operating state as the input variable of the VAE, and obtain the latent variable through non-linear dimensionality reduction, that is, the non-linear degradation feature Z = (z 1 , z 2 ,..., z j ) of the aero-engine; and optimize the learning process of the VAE by maximizing the objective function. The formula expression of the objective function is

[0029] logp(y) = log∫p(z)p(y|z);

[0030] where, p(y) and p(z) respectively represent the probability distributions of the screened multi-source data of the operating state and the non-linear degradation feature; p(y|z) is the conditional probability distribution, indicating the distribution of y given z;

[0031] S32. By introducing the posterior distribution q(z|y) to approximate p(y|z), the objective function is transformed into:

[0032]

[0033] Among them, is the lower bound of evidence, and D KL (q(z|ys)||p(z|y)) is the KL divergence;

[0034] S33. According to the fact that the KL divergence is always not less than 0, the maximization of the objective function is transformed into the maximization of the lower bound of evidence The formula of the lower bound of evidence is expressed as:

[0035]

[0036] Among them, E q(z|y) [logp(y|z)] is the reconstruction loss;

[0037] By maximizing the lower bound of evidence, the latent variable close to the distribution of the input variable is obtained.

[0038] Furthermore, step S4 specifically includes:

[0039] S41. Based on the filtered multi-source data of the operating state, calculate the mean value μ of the operating state data collected by the sensor j on the aero-engine i during the complete operating moment; the formula is expressed as:

[0040]

[0041] S42. Calculate the covariance matrix ψ, and the formula is expressed as:

[0042]

[0043] Among them, is the transpose matrix of the matrix ;

[0044] S43. Perform sparse principal component analysis on the covariance matrix, and obtain the sparse principal component direction vector s by optimizing the following objective function:

[0045] maxs T ψs - α||s|| 1 ;

[0046] Among them, α is the sparse regularization parameter that controls the degree of sparsity, s is the sparse principal component direction vector, is the unit vector, and s T represents the transpose vector of the sparse principal component direction vector, and ||s|| 1 represents the L1 norm of s, which is used to introduce sparsity;

[0047] S44. Perform a linear transformation on the multi-source operation status data after reliability screening using the obtained sparse principal component vectors to obtain the linear degradation feature Q = sY = (q 1 , q 2 ,..., q j ).

[0048] Furthermore, step S5 specifically includes:

[0049] S51. Set the dynamic fusion weight of the degradation feature as γ(t) = [γ 1 (k), γ 2 (k)];

[0050] where γ 1 (k) and γ 2 (k) are the fusion weights of the linear degradation feature and the non-linear degradation feature respectively; representing the importance and weight allocation of the corresponding degradation feature, which changes dynamically with time k;

[0051] S52. In the initial state, it is considered that the importance of the linear degradation feature and the non-linear degradation feature is equal, so the weights are set as:

[0052] γ 1 (1) = γ 2 (1) = 0.5;

[0053] S53. Calculate the change rates of the linear degradation feature and the non-linear degradation feature at any time k; the formula is expressed as:

[0054]

[0055] where Q(k) and Z(k) represent the linear degradation feature and the non-linear degradation feature at time k respectively, and Q(k - 1) and Z(k - 1) represent the linear degradation feature and the non-linear degradation feature at time k - 1 respectively; Δk = k - (k - 1);

[0056] S54. Then perform normalization processing on the feature change rates to obtain:

[0057]

[0058] S55. According to the normalized change rates, use the exponentially weighted moving average algorithm to obtain the dynamic fusion weights; the formula is expressed as:

[0059] γ 1 (k) = β·r SPCA,norm (k) + (1 - β)·γ 1 (k - 1)

[0060] γ 2 (k) = β·r VAE,norm(k) + (1 - β)·γ 2 (k - 1);

[0061] Where β is the weight adjustment rate factor, which controls the response sensitivity of weight update to the current feature change;

[0062] S56. Based on the degradation features obtained in steps S3 and S4 and the dynamic fusion weights in step S55, construct a composite health index H, and the formula is expressed as:

[0063] H = γ 1 (k)Q + γ 2 (k)Z;

[0064] Where Q and Z are the linear degradation feature and the non - linear degradation feature respectively.

[0065] In addition, the present invention also provides a device, which includes a memory and a processor, where:

[0066] The memory is used to store a computer program that can run on the processor;

[0067] The processor is used to execute the method for modeling the aero - engine health index by fusing linear and non - linear features as described above when running the computer program.

[0068] The present invention also provides a computer - readable storage medium, which stores computer instructions for causing the processor to execute the method for modeling the aero - engine health index by fusing linear and non - linear features as described above.

[0069] Based on the above technical solutions, the present invention has at least the following beneficial effects:

[0070] 1. It fuses multi - source data of the operating state of the aero - engine, fully excavates the linear and non - linear features collected by multi - source sensors, and helps to improve the accuracy of health state assessment;

[0071] 2. It combines the advantages of non - linear methods and linear methods, and improves the robustness of data processing and the efficiency and interpretability of feature expression;

[0072] 3. The aero - engine health index constructed by the method proposed by the present invention has higher quality, can more intuitively reflect the degradation trend of the aero - engine, and provides strong support for subsequent remaining life prediction and maintenance decision - making. Description of the Drawings

[0073] The drawings described herein are used to provide a further understanding of the present application, and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation to the present application. In the drawings:

[0074] Figure 1 Flow chart for modeling the health indicators of an aero - engine by the method proposed in the present invention;

[0075] Figure 2 Trend graph of the operating state data collected by 21 sensors on a single randomly selected aero - engine in the training set over time in an embodiment of the present invention;

[0076] Figure 3 Trend graph of the composite health indicators obtained by different methods for four test aero - engines involved in an embodiment of the present invention over time. Detailed implementation manners

[0077] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present 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 only used to explain the present invention and are not used to limit the present invention.

[0078] Although the steps in the present invention are numbered, they are not used to limit the order of the steps, unless the order of the steps is clearly stated or the execution of a certain step requires other steps as a basis. Otherwise, the relative order of the steps can be adjusted. It can be understood that the term "and / or" used herein relates to and encompasses any and all possible combinations of one or more of the associated listed items.

[0079] As Figures 1 - 3 shown, the present invention proposes a method for modeling the health indicators of an aero - engine by fusing linear and non - linear features; as Figure 1 shown, it specifically includes the following steps:

[0080] S1. Arrange a multi - source sensor detection system to collect multi - source data on the operating state of the aero - engine;

[0081] As a preferred implementation manner, step S1 specifically includes:

[0082] S11. Install and connect multiple sensors on the aero - engine to measure different physical quantities;

[0083] The aero - engine involved in this embodiment mainly includes a fan, a combustion chamber, a low - pressure turbine, a low - pressure compressor, a high - pressure compressor, a high - pressure turbine and a nozzle; at the initial stage of component degradation, the engine operates normally. As the service time of the engine continues, the engine performance gradually degrades until the entire engine fails completely. Therefore, in this embodiment, 21 sensors such as pressure, temperature, and speed are installed inside the engine to measure the degradation state of the performance of the aero - engine in each dimension over time.

[0084] S12. Configure data transmission mode, sampling frequency, and data storage format parameters;

[0085] S13. After determining that the sensor is in a normal working state, collect multi-source data on the operating state of the aero-engine, and its formula is expressed as:

[0086]

[0087] Among them, A i is the multi-source data on the operating state of aero-engine i; is the operating state data collected by sensor j on aero-engine i at operating time k;

[0088] In addition, it should be noted that represents the operating state data collected by sensor j on aero-engine i at all operating times.

[0089] S2. Perform normalization processing on the collected multi-source data on the operating state, and then use the Pearson coefficient for reliability screening;

[0090] As a preferred implementation method, step S2 specifically includes:

[0091] S21. Perform normalization processing on the collected multi-source data on the operating state of the aero-engine to avoid the influence of different sensor dimensions; the normalized multi-source data on the operating state is expressed as:

[0092]

[0093] Among them, U i is the normalized multi-source data on the operating state of aero-engine i; is the normalized operating state data collected by sensor j on aero-engine i at operating time k; u i,j The specific form of i,j ;

[0094] S22. Use the Pearson coefficient to perform reliability screening on the normalized multi-source data on the operating state, and eliminate the operating state data collected by sensors with small Pearson coefficients, that is, weak correlations. The screened multi-source data on the operating state is recorded as Y = (y i,1 , y i,2 ,..., y i,j ); The calculation formula of the Pearson coefficient is:

[0095]

[0096] Among them, is the mean value of the sum of historical moment values of aero-engine i, is the k-th operating moment of the aero-engine i, is the mean value of the normalized operating state data collected by the sensor j on the engine i; K is the maximum operating moment of the aero-engine i.

[0097] In this embodiment, the Pearson coefficient is first used to measure the correlation between the state data collected by different sensors and the engine operating moment, that is, the sensor data with weak correlation with the degradation of the engine health state is screened out, ensuring the reliability of the subsequent extracted non-linear features and linear features.

[0098] S3. Use the variational autoencoder VAE to perform data fusion on the remaining multi-source operating state data after screening to obtain non-linear degradation features;

[0099] As a preferred implementation manner, step S3 specifically includes:

[0100] S31. Use the remaining multi-source operating state data after screening as the input variable of the VAE, and obtain the latent variable through non-linear dimensionality reduction processing, that is, the non-linear degradation feature Z=(z 1 ,z 2 ,...,z j ) of the aero-engine; and optimize the learning process of the VAE by maximizing the objective function; the formula expression of the objective function is

[0101] logp(y)=log∫p(z)p(y|z);

[0102] Among them, p(y) and p(z) respectively represent the probability distributions of the remaining multi-source operating state data after screening and the non-linear degradation features; p(y|z) is the conditional probability distribution, indicating the distribution of y given z;

[0103] S32. By introducing the posterior distribution q(z|y) to approximate p(y|z), the objective function is converted to:

[0104]

[0105] Among them, is the evidence lower bound, D KL (q(z|ys)||p(z|y)) is the KL divergence, used to measure the difference between the probability distributions of the input variable and the latent variable;

[0106] S33. According to the fact that the KL divergence is always not less than 0, maximizing the objective function is converted to maximizing the evidence lower bound The formula representation of the evidence lower bound is:

[0107]

[0108] Among them, E q(z|y)[logp(y|z)] is the reconstruction loss; it is used to measure the similarity between the input variable and the latent variable;

[0109] By maximizing the evidence lower bound, the latent variable close to the distribution of the input variable is obtained.

[0110] S4. Perform data fusion on the remaining multi-source operation status data after screening through the sparse principal component analysis (SPCA) method to obtain linear degradation features;

[0111] As a preferred implementation, step S4 specifically includes:

[0112] S41. Based on the screened multi-source operation status data, calculate the mean value μ of the operation status data collected by the sensor j on the aero-engine i during the complete operation time; the formula is expressed as:

[0113]

[0114] S42. Calculate the covariance matrix ψ, and the formula is expressed as:

[0115]

[0116] where is the transpose matrix of the matrix ;

[0117] S43. Perform sparse principal component analysis on the covariance matrix, and obtain the sparse principal component direction vector s by optimizing the following objective function:

[0118] maxs T ψs - α||s|| 1 ;

[0119] where α is the sparse regularization parameter that controls the sparsity degree, s is the sparse principal component direction vector, is the unit vector, s T represents the transpose vector of the sparse principal component direction vector, and ||s|| 1 represents the L1 norm of s, which is used to introduce sparsity;

[0120] S44. Use the obtained sparse principal component vector to perform a linear transformation on the multi-source operation status data after reliability screening to obtain the linear degradation feature Q = sY = (q 1 , q 2 ,..., q j ).

[0121] S5. According to the obtained non-linear and linear degradation features, design dynamic fusion weights, and construct a composite health index by combining the weighted moving average optimization algorithm;

[0122] As a preferred implementation, step S5 specifically includes:

[0123] S51. Set the dynamic fusion weight of the degradation features as γ(t) = [γ 1 (k), γ 2 (k)];

[0124] Among them, γ 1 (k) and γ 2 (k) are the fusion weights of the linear degradation feature and the non - linear degradation feature respectively; they represent the importance and weight distribution of the corresponding degradation features, and change dynamically with the k - th moment;

[0125] S52. In the initial state, it is considered that the importance of the linear degradation feature and the non - linear degradation feature is equal, and the weights are set as:

[0126] γ 1 (1) = γ 2 (1) = 0.5;

[0127] S53. Calculate the change rates of the linear degradation feature and the non - linear degradation feature at any k - th moment; the formula is expressed as:

[0128]

[0129] Among them, Q(k) and Z(k) represent the linear degradation feature and the non - linear degradation feature at the k - th moment respectively, and Q(k - 1) and Z(k - 1) represent the linear degradation feature and the non - linear degradation feature at the (k - 1) - th moment respectively; Δk = k - (k - 1);

[0130] S54. Then normalize the feature change rates to obtain:

[0131]

[0132] S55. According to the normalized change rates, use the exponentially weighted moving average algorithm to obtain the dynamic fusion weights; the formula is expressed as:

[0133] γ 1 (k) = β·r SPCA,norm (k)+(1 - β)·γ 1 (k - 1)

[0134] γ 2 (k) = β·r VAE,norm (k)+(1 - β)·γ 2 (k - 1);

[0135] Among them, β is the weight adjustment rate factor, which controls the response sensitivity of the weight update to the current feature change;

[0136] S56. Based on the degradation features obtained in steps S3 and S4 and the dynamic fusion weights in step S55, construct a composite health index H, which is expressed by the formula:

[0137] H = γ 1 (k)Q + γ 2 (k)Z;

[0138] Among them, Q and Z are the linear degradation feature and the non - linear degradation feature respectively.

[0139] So far, the specific process of the method proposed by the present invention has been described. Next, the superiority of the method proposed by the present invention will be verified through specific experimental examples. In this embodiment, to verify the performance of the aero - engine health index modeling method that fuses linear and non - linear features proposed by the present invention, this method is executed on the publicly available aero - engine degradation dataset provided by NASA. In the offline stage, 100 groups of complete time series from the start of operation to the end of failure included in the "train_FD001.txt" file in the dataset are used. Among them, there is sensor data that is transmitted to the central processing unit in real - time based on wireless communication technology. The sampling frequency is set to 1 time per cycle to ensure efficient monitoring of the engine operation state. The data storage format adopts the standard text file (.txt) format. Each row record contains the engine number, timestamp, operation condition parameters, and 21 sensor data for subsequent analysis and processing. When verifying, 100 groups of state monitoring sequences of incomplete test engines included in the "test_FD001.txt" file in the dataset are used.

[0140] In this embodiment, first, pre - process the original data of the engine operation state collected by the sensor according to the reliability screening method in step S2; the operation state data collected by 21 sensors in the training set is as Figure 2 shown. It can be seen that the values of the operation state data collected by some sensors change significantly over time, reflecting the degradation of the aero - engine performance over time; while the sensor data with no obvious change in the figure is regarded as having weak correlation and is screened out and does not participate in subsequent feature extraction. Through the reliability screening in step S2 of this application, the degradation features collected by sensors that are as strongly correlated as possible with the aero - engine performance degradation are selected as much as possible, which helps the rationality and reliability of constructing the composite health index subsequently.

[0141] Then, according to steps S3 and S4, extract the non - linear degradation feature and the linear degradation feature from the remaining multi - source operation state data after screening through the variational auto - encoder VAE and the sparse principal component analysis SPCA methods respectively, and construct a composite health index according to step S5, and use the exponentially weighted moving average algorithm to optimize the fusion coefficient γ = [γ 1 , γ 2 .

[0142] After the training is completed, the expressions of the composite health index and the dynamic fusion weight are obtained. At this time, the incomplete state detection data in the "test_FD001.txt" file is tested according to the above steps; the test results are as Figure 3 shown, which describe the health index values obtained by using the method proposed by the invention and using VAE or SPCA alone. From Figure 3 it can be seen that compared with constructing the degradation feature vector by using VAE or SPCA alone, the composite health index obtained by the method proposed by the invention has smaller fluctuations over time, and the overall trend is stable. The index values calculated at the initial and final stages of the engine operation are not too small or too large, which reflects that the method proposed by the invention has better robustness. In addition, after calculating the Pearson coefficient, the highest Pearson coefficient between the single sensor data in the original multi-source data of the aero-engine without processing and the operation time is 0.811; after constructing the degradation feature vector by using VAE or SPCA alone and then calculating the Pearson coefficient, they are 0.823 and 0.837 respectively, while the Pearson coefficient between the degradation feature constructed by the method proposed by the invention and the operation time is 0.851, which reflects that the designed composite health index of the invention has a stronger correlation with the health state of the engine. Therefore, the composite health index modeling method proposed by the invention is reliable.

[0143] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to include all changes falling within the meaning and scope of the equivalent elements of the claims in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0144] In addition, it should be understood that although this specification is described according to the embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An aircraft engine health index modeling method integrating linear and nonlinear characteristics, characterized in that: The specific steps include: S1. Arrange a multi-source sensor detection system to collect multi-source data on the operating status of the aircraft engine; S2. Normalize the collected multi-source data of operation status, and then use the Pearson coefficient to perform reliability screening; S3, using variational autoencoder VAE to fuse the remaining multi-source data of the operating status after screening to obtain nonlinear degradation features; S4. The sparse principal component analysis (SPCA) method is used to fuse the multi-source data of the operating status remaining after screening to obtain linear degradation characteristics; S5. Based on the obtained nonlinear and linear degradation characteristics, dynamic fusion weights are designed, and a composite health index is constructed in combination with the weighted moving average optimization algorithm.

2. The aircraft engine health index modeling method integrating linear and nonlinear characteristics according to claim 1 is characterized in that: Step S1 specifically includes: S11. Install and connect multiple sensors measuring different physical quantities on the aircraft engine; S12, configure data transmission mode, sampling frequency, and data storage format parameters; S13. After determining that the sensor is in a normal working state, collect multi-source data of the operating state of the aircraft engine, and the formula is expressed as follows: Among them, A i is the multi-source data of the operating status of aircraft engine i; It is the operating status data collected by sensor j on aircraft engine i at operating time k.

3. The aircraft engine health index modeling method integrating linear and nonlinear characteristics according to claim 2 is characterized in that: Step S2 specifically includes: S21. Normalize the collected multi-source data of the operating status of the aircraft engine to avoid the influence of different sensor dimensions; the normalized multi-source data of the operating status is expressed as: Among them, U i is the normalized multi-source data of the operating status of aircraft engine i; is the normalized operating status data collected by sensor j on aircraft engine i at operating time k; S22. The normalized multi-source data of the running status are screened for reliability using the Pearson coefficient, and the running status data collected by sensors with small Pearson coefficients, i.e., weak correlations, are eliminated. The screened multi-source data of the running status is recorded as Y=(y i,1 ,y i,2 ,...,y i,j ); The calculation formula of Pearson coefficient is: in, is the mean of the sum of the historical values ​​of aircraft engine i, is the kth operating moment of aircraft engine i, is the mean of the normalized operating status data collected by sensor j on engine i; K is the maximum operating time of aircraft engine i.

4. The aircraft engine health index modeling method integrating linear and nonlinear characteristics according to claim 1 is characterized in that: Step S3 specifically includes: S31. Using the filtered multi-source data of operating status as the input variable of VAE, the latent variable is obtained through nonlinear dimensionality reduction processing, that is, the nonlinear degradation characteristic Z of the aircraft engine = (z1, z2, ..., z j ); and optimize the learning process of VAE by maximizing the objective function; the formula of the objective function is expressed as logp(y)=log∫p(z)p(y|z); Among them, p(y) and p(z) represent the probability distribution of the filtered multi-source data of the operating status and the nonlinear degradation characteristics, respectively; p(y|z) is the conditional probability distribution, which represents the distribution of y when z is given; S32. By introducing the posterior distribution q(z|y) to approximate p(y|z), the objective function is converted to: in, is the lower bound of evidence, D KL (q(z|ys)||p(z|y)) is the KL divergence; S33. According to the KL divergence is always not less than 0, the maximization objective function is transformed into maximizing the lower bound of evidence The formula for the lower bound of evidence is expressed as: Among them, E q(z|y) [logp(y|z)] is the reconstruction loss; By maximizing the lower bound of evidence, we can obtain latent variables that are close to the distribution of input variables.

5. The aircraft engine health index modeling method integrating linear and nonlinear characteristics according to claim 1 is characterized in that: Step S4 specifically includes: S41. Based on the filtered multi-source operating status data, calculate the mean μ of the operating status data collected by sensor j on aircraft engine i during the complete operating time; the formula is expressed as: S42, calculate the covariance matrix ψ, the formula is expressed as: in, For the matrix The transposed matrix of S43, perform sparse principal component analysis on the covariance matrix, and obtain the sparse principal component direction vector s by optimizing the following objective function: maxs T ψs-α||s||1; Among them, α is the sparse regularization parameter, which controls the degree of sparsity, s is the sparse principal component direction vector, is a unit vector, and s T represents the transposed vector of the sparse principal component direction vector, ||s||1 represents the L1 norm of s, which is used to introduce sparsity; S44, using the obtained sparse principal component vector to perform linear transformation on the multi-source data of the operating status after reliability screening, to obtain the linear degradation feature Q = sY = (q1, q2, ..., q j ).

6. The aircraft engine health index modeling method integrating linear and nonlinear characteristics according to claim 1 is characterized in that: Step S5 specifically includes: S51, setting the dynamic fusion weight of the degenerate feature to γ(t)=[γ1(k),γ2(k)]; Among them, γ1(k) and γ2(k) are the fusion weights of linear degradation features and nonlinear degradation features respectively; they represent the importance and weight distribution of the corresponding degradation features, and change dynamically with time k; S52. In the initial state, it is considered that the importance of linear degradation features and nonlinear degradation features is equal, and the weight is set as: γ1(1)=γ2(1)=0.5; S53. Calculate the rate of change of the linear degradation characteristics and the nonlinear degradation characteristics for any k moment; the formula is expressed as: Where, Q(k) and Z(k) represent the linear degradation characteristics and nonlinear degradation characteristics at time k, respectively; Q(k-1) and Z(k-1) represent the linear degradation characteristics and nonlinear degradation characteristics at time k-1, respectively; Δk = k-(k-1); S54, normalize the feature change rate and obtain: S55. According to the normalized rate of change, the exponentially weighted moving average algorithm is used to obtain the dynamic fusion weight; the formula is expressed as: Among them, β is the weight adjustment rate factor, which controls the sensitivity of weight update to the current feature changes; S56, based on the degradation characteristics obtained in steps S3 and S4 and the dynamic fusion weights in step S55, a composite health index H is constructed, and the formula is expressed as: H = γ1(k)Q + γ2(k)Z; Among them, Q and Z are linear degradation characteristics and nonlinear degradation characteristics respectively.

7. A device, characterized in that: The device comprises a memory and a processor, wherein: A memory for storing computer programs that can be run on the processor; A processor is used to execute the aircraft engine health index modeling method integrating linear and nonlinear characteristics as described in any one of claims 1 to 6 when running the computer program.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the aircraft engine health index modeling method integrating linear and nonlinear characteristics according to any one of claims 1 to 6 when executed.

Images