Aero-engine gas path parameter deviation value preprocessing method

Abstract

The present application relates to the technical field of aero-engine gas path parameter processing, in particular to a kind of aero-engine gas path parameter deviation value preprocessing method, first, using fingerprint map to correct the deviation of gas path parameter caused by non-component performance degradation in gas path parameter deviation value offset;Then, using Grubbs criterion to remove gross error in short-term gas path parameter deviation value that meets the requirement of independent and identically distributed by equal-length overlapping sliding window division;Then, aiming at the problems such as mode coupling and insufficient IMF signal based on correlation coefficient selection that EMD decomposition used by general noise reduction method may encounter, EEMD decomposition is introduced, and single-sample T test is used to distinguish high-frequency and low-frequency IMF signal, by only using SG filter to reduce noise for high-frequency IMF, the problem that long-term trend signal is damaged due to data filtering and noise reduction is avoided;Finally, the superiority of the data preprocessing method proposed in the present application in gas path parameter deviation value preprocessing is verified by comparison experiment.

Description

Technical fields:

[0001] This invention relates to the field of aero-engine gas path parameter processing technology, specifically a preprocessing method for aero-engine gas path parameter deviation values ​​that can reduce data noise and decrease subsequent data processing errors. Background technology:

[0002] Statistics show that 88% of aviation accidents are caused by failures in aircraft subsystems or propulsion systems. This results in maintenance costs accounting for 50% of the total life-cycle costs of aviation equipment, with engine maintenance costs accounting for 39% of these costs. This poses a significant challenge to China's capital-intensive aviation industry.

[0003] As a core research area of ​​engine health management, engine airflow performance assessment is fundamental to engine operation and maintenance management, and it has positive implications for monitoring engine service status, defining maintenance scope, and predicting long-term lifespan. Airflow performance analysis primarily involves analyzing engine airflow operation monitoring parameters through mechanistic, data, or multi-source information fusion models to determine the overall engine or unit's health status. Mechanism-based analysis methods mainly include linear and nonlinear models, such as the airborne adaptive model for aero-engines based on linear Kalman filters to estimate the deviation between state parameters and model parameters, and airflow fault diagnosis systems based on nonlinear extended Kalman filters. Data-based airflow performance analysis methods mainly utilize expert systems, deep learning, and machine learning, such as fuzzy logic expert reasoning systems based on knowledge and rules, fault classification systems based on backpropagation neural networks, and performance diagnosis systems based on small-sample support vector machines. Multi-source information fusion models rely on the fusion of multidisciplinary knowledge and data for performance analysis, such as engine health status assessment based on Bayesian networks or DS evidence theory.

[0004] Mechanical fault diagnosis primarily involves analyzing vibration signals and lubricating oil metal debris from engine bearings, gears, rotors, and accessories to understand the deterioration and damage status of engine components. Vibration signal analysis mainly includes bearing signal analysis based on Hilbert transform, wavelet decomposition, or empirical mode decomposition; gear signal analysis based on health indicator (HI) signals; and rotor signal analysis based on higher-order spectral methods. Lubricating oil analysis mainly includes offline lubricating oil monitoring and online metal detection.

[0005] As a characteristic parameter of aero-engine airflow performance, the deviation value of airflow parameters is affected by the performance of engine airflow components under coupled alternating operating conditions and the external operating environment. The performance of airflow components is further affected by the engine control system's adjustments and their own performance degradation. Therefore, it is characterized by high short-term dispersion, large coefficient of variation, and slow long-term fluctuations. Thus, if airflow parameter deviation values ​​are to be used as a data basis for evaluating the performance of multiple engine airflow components, data cleaning is necessary before the evaluation. This cleaning should correct data offsets caused by non-component performance degradation, remove gross errors caused by non-operating-condition factors such as unstable airborne sensor operation, and reduce data noise. Summary of the Invention:

[0006] This invention addresses the shortcomings and deficiencies of existing technologies by proposing a preprocessing method for aero-engine air path parameter deviations that can reduce data noise and minimize subsequent data processing errors.

[0007] This invention achieves its purpose through the following measures:

[0008] A method for preprocessing deviation values ​​of aero-engine air path parameters, characterized by comprising the following steps:

[0009] Step 1: For the deviation value of the gas path parameters The correction formula based on the fingerprint image is as follows:

[0010]

[0011] In the formula —The uncorrected deviation value of the original gas path parameters at time i;

[0012] x ij —The unit opening angle or unit bleed air leakage of VSV and VBV at time i;

[0013] w j —x in the fingerprint image j The offset caused by deviations in gas path parameters;

[0014] Step 2: Use the Grubbs criterion to remove gross errors from the grouped gas path parameter deviations;

[0015] Step 3: Use EEMD-SG to filter and reduce noise in the gas path parameter deviation value D after gross error removal.

[0016] The specific steps are as follows:

[0017] Step 3-1: Decompose the gas path parameter deviation values ​​based on EEMD;

[0018] Step 3-2: Use a one-sample t-test to perform a zero-mean test on the cumulative sum of IMF components to distinguish between high-frequency and low-frequency IMFs. Those that pass the test are high-frequency IMFs, and those that fail are low-frequency IMFs. The specific steps are as follows:

[0019] Step 3-2-1: Calculate the cumulative IMF component I according to formula (12). s :

[0020]

[0021] Step 3-2-2: Establish the test hypothesis and determine the significance level: Let the test mean μ = 0, and the null hypothesis H0: μ Is =μ, Alternative hypothesis H1: μ Is ≠μ, significance level α=0.05, where μ Is For the cumulative IMF component I s The mean;

[0022] Step 3-2-3: Select the test method and calculate the test statistic: Calculate the statistic using the one-sample T-test formula (13) and perform a two-tailed test:

[0023]

[0024] In the formula, k——I s The number of samples;

[0025] σ Is —I s Standard deviation;

[0026] Step 3-2-4: Determine the P-value and draw inferences: Calculate t according to equation (13). α,k-1 The value is determined by consulting the T-distribution table. If P > α, then H0 is accepted, and the IMF component I is accumulated. s Pass the zero-mean t-test; if P < α, then accept H1 and accumulate the IMF component I. s The zero-mean t-test was not passed when the cumulative IMF component that passed the test was I. d At that time, the IMF components with j < d are high-frequency signals, and the IMF components with j > d are low-frequency signals; Step 4: Smooth the high-frequency IMF signal using the Savitzky-Golay filter. The specific steps are as follows:

[0027] Step 4-1: For high-frequency IMF components I h Within an overlapping sliding window of equal length, there are 2m+1 consecutive values ​​x. i , i∈(-m,m), and fit the data by constructing a k-th order polynomial (k≤2m+1) according to equation (14).

[0028]

[0029] In the formula a j —Undetermined coefficients of the polynomial;

[0030] Step 4-2: Determine the undetermined coefficients a of the polynomial j The residual E between the fitted data and the original data can be obtained by least squares fitting using equation (15):

[0031]

[0032] Step 4-3: To minimize the residual E, we need to set the reciprocal of E with respect to the undetermined coefficients of each polynomial to 0.

[0033]

[0034] Solving for:

[0035]

[0036] The center point estimate of the window is calculated using the fitted polynomial, and then the high-frequency intrinsic mode components I at any time step are traversed using an equal-length overlapping sliding window. h The signal smoothing process can be completed to obtain the reconstructed high-frequency IMF signal I. h ′;

[0037] Step 5: Reconstruct the deviation value of the gas path parameters after noise reduction: According to formula (18), the high-frequency IMF component I obtained after smoothing by the SG filter is... h Low-frequency IMF component I l By summing the residual component R, the deviation value D′ of the reconstructed gas path parameters can be obtained.

[0038]

[0039] In the formula I i — High-frequency IMF components processed by the SG filter;

[0040] I i —Low-frequency IMF components not processed by the SG filter;

[0041] R is the residual component obtained from EEMD decomposition.

[0042] The specific steps of step 2 of this invention are as follows:

[0043] Step 2-1: For a set of corrected gas path parameter deviations that follow a Gaussian distribution The mean μ and standard deviation σ of this set of data can be calculated using equations (2) and (3).

[0044]

[0045] Step 2-2: This set of data Arrange in ascending order according to formula (4): d (1) <d (2) <…<d (n) (4);

[0046] Steps 2-3: Calculate the sorted data according to formula (5). The corresponding statistic g (i) :

[0047] If data point d (i) The corresponding statistic g (i) >g (0) If (n, α), then this point represents a gross error and needs to be removed from the data set, where g (0) (n,α) is the critical value of the statistic when the number of repeated measurements is n and the confidence probability is α, which can be obtained by consulting the Grubbs critical value table;

[0048] Steps 2-4: To ensure data integrity, the data d to be removed... i The data is completed using interpolation method (6):

[0049] Step 3-1 of this invention specifically includes the following steps:

[0050] Step 3-1-1: Set the number of repetitions m of the superimposed Gaussian white noise and the amplitude α of the white noise relative to the standard deviation of the original signal. Step 3-1-2: Superimpose Gaussian white noise N with amplitude α into the gas path parameter deviation value D after gross error removal according to formula (8). i The new signal obtained is denoted as Step 3-1-3: Use EMD to analyze the new signal The decomposition is performed, and it is decomposed into n intrinsic mode components I according to equation (9). i,j and one residual component R i : Step 3-1-4: Repeat steps 3-1-2 and 3-1-3 m times, and calculate the mean values ​​of each IMF component and residual component obtained according to equations (10) and (11) as the intrinsic mode components I after decomposition of the gas path parameter deviation value D. j and residual components R:

[0051]

[0052] Data noise is mainly contained in the high-frequency intrinsic mode components I. hIn the above, the noise component is considered, while the signal trend is mainly reflected in the low-frequency intrinsic mode component I. l In the residual component R, it can be regarded as an evolutionary component, so data denoising mainly depends on the high-frequency intrinsic mode component I. h The processing.

[0053] Compared with existing technologies, this invention effectively eliminates the influence of engine unsteady conditions on the deviation values ​​of air path parameters by using Grubbs' criterion and EEMD-SG to preprocess the deviation values ​​of air path parameters, thereby removing gross errors in the data and reducing data noise. Attached image description:

[0054] Appendix Figure 1 It is a fingerprint image of a CFM56-3 / -5A / -5B / -7B aircraft engine.

[0055] Appendix Figure 2 This is a schematic diagram illustrating the principle of calculating the deviation value of gas path parameters.

[0056] Appendix Figure 3 These are the deviation values ​​of the gas path parameters before and after the correction.

[0057] Appendix Figure 4 This is a schematic diagram of the short-term data distribution of the engine.

[0058] Appendix Figure 5 High-frequency intrinsic mode components I h and low-frequency intrinsic mode components I l Schematic diagram.

[0059] Appendix Figure 6 This is a schematic diagram of the cumulative IMF signals separated by the T-test.

[0060] Appendix Figure 7 It is an analog signal Y ps and noisy signal Y ns Schematic diagram.

[0061] Appendix Figure 8 This is a schematic diagram of the signal after noise reduction for each experimental method in the embodiments of the present invention.

[0062] Appendix Figure 9 It is ΔEGT after removing gross errors using the Grubbs criterion.

[0063] Appendix Figure 10 It is the IMF component after ΔEGT is decomposed by EEMD.

[0064] Appendix Figure 11 It is the IMF component of ΔEGT after SG filtering.

[0065] Appendix Figure 12 This is a schematic diagram of the deviation values ​​of the pre-processed gas path parameters. Detailed implementation method:

[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0067] Example 1:

[0068] This example first corrects the data offset caused by the performance degradation of non-gas path components by using the fingerprint diagram to indicate the interference. Then, the data divided by equal-length overlapping sliding windows is subjected to gross error removal using the Grubbs criterion. Next, the gas path parameter deviation is decomposed into multiple intrinsic mode functions (IMFs) and a residual component using ensemble empirical mode decomposition (EEMD). Then, a one-sample t-test is used to distinguish between high-frequency and low-frequency IMFs, treating low-frequency IMFs as evolutionary components and high-frequency IMFs as noisy components. The noisy component is filtered using an SG filter. Finally, the gas path parameter deviation is reconstructed by combining the evolutionary component, the residual component, and the filtered noisy component, thus achieving the gas path parameter deviation preprocessing required for the performance evaluation of multiple components in the engine gas path.

[0069] The fingerprint diagram described in this example is used to indicate the numerical deviation of each air path parameter due to changes in the engine's air path operating state when certain control operations or performance changes of air path components occur under certain flight conditions and operating states of a normally on-wing engine. Chapter 5, "Cruise Sensitivities," of the Engine Diagnostics Trend Interpretation manual issued by the manufacturer to the operations and maintenance management department provides fingerprint diagrams for the status monitoring of various CFM engine models. The fingerprint diagram for the CFM56-3 / -5A / -5B / -7B aero engines is shown below. Figure 1 As shown.

[0070] This figure shows the deviation of air path parameters of the CFM56-3 / -5A / -5B / -7B aero engine under the following conditions: flight altitude of 35,000 feet, Mach number of 0.76, International Standard Atmosphere (ISA) operating environment, and constant low-speed shaft speed N1. These deviations are relative to the values ​​when no changes occur in the air path operating conditions.

[0071] The air path parameter deviation value represents the parameter offset between the original air path parameter values ​​calculated from the engine operating environment and control parameters recorded by onboard sensors and converted to the actual operating conditions specified in the fingerprint diagram, and the theoretical operating air path parameters calculated based on the baseline model simplified from the engine OEM's thermodynamic theoretical model. Figure 2 As shown.

[0072] like Figure 1 As shown, the main deviation values ​​of the gas path parameters of aero-engine are as follows: (1) Deviation of exhaust gas temperature (ΔEGT); (2) Deviation of fuel flow (ΔFF); (3) Deviation of high-pressure rotor speed (ΔN2).

[0073] like Figure 1 As shown, the factors affecting the deviation values ​​of the air passage parameters mainly include two parts: ① the opening and closing angles of VSV and VBV and the amount of bleed air leakage, and ② the efficiency degradation of engine air passage components. This invention focuses on the performance evaluation of engine air passage components, therefore it is necessary to correct the deviation values ​​of the air passage parameters based on the fingerprint diagram to eliminate the data offset caused by non-component performance degradation factors.

[0074] For gas path parameter deviation value The correction formula based on the fingerprint image is as shown in equation (1):

[0075] In the formula —The uncorrected deviation value of the original gas path parameters at time i;

[0076] x ij —The unit opening angle or unit bleed air leakage of VSV and VBV at time i;

[0077] w j —x in the fingerprint image j The offset caused by deviations in the gas path parameters.

[0078] Table 1 shows 3959 points of deviation values ​​for the air path parameters throughout the entire service life of an engine from the start of its service to its first major overhaul. The air path deviation values ​​were corrected using the formula above. The data before and after correction are as follows: Figure 3 As shown.

[0079] Table 1. Operational data required to correct gas path parameter deviations based on fingerprint diagrams.

[0080]

[0081]

[0082] Before the correction, the deviation of the gas path parameters of the engine during its service life to the first major overhaul showed a slow upward trend due to the changes in the operating status of the engine and its components and the performance degradation. That is, due to the aging of the engine's service condition, the reliability and safety decreased, and the deviation of the gas path parameters also showed a slow increasing trend.

[0083] After data correction, it is evident that the opening angles of VSV and VBV, as well as the engine bleed air volume, do not consistently influence the disturbances to ΔEGT, ΔFF, and ΔN2. The corrected ΔEGT and ΔFF data are more concentrated, while ΔN2 shows a significant shift from the baseline. Therefore, it is essential to correct the deviations in the airflow parameters based on the fingerprint diagram before evaluating the performance of engine airflow components.

[0084] like Figure 3 As shown, even after correcting for data offsets in the air path parameter deviation values ​​that are not caused by component performance degradation using fingerprint diagrams, a significant amount of data noise remains in the air path parameter deviation values ​​due to changes in the engine operating environment and control parameters. This noise severely affects the stability of the performance evaluation of engine air path components. Therefore, before using air path parameter deviation values ​​for engine air path component performance evaluation, it is necessary to perform data preprocessing on the air path parameter deviation values ​​corrected by fingerprint diagrams to eliminate gross errors and random noise in the data.

[0085] Gross errors are defined as unreliable data where, under certain measurement conditions, the measured value relative to the actual value significantly exceeds the expected range. These errors can be removed using physical and statistical discrimination methods. Due to the prior nature of physical discrimination methods, statistical discrimination methods are used here for gross error removal. For example... Figure 3 As shown, overall, the corrected deviations in the air path parameters of the entire lifespan aero-engine are clearly non-steady-state time-series data that do not conform to independent and identically distributed distributions. Their values ​​change over time due to variations in engine operating conditions and component performance degradation. However, for short-term data within 7 days, seasonal temperature changes have a relatively small impact on engine environmental conditions; flight plans are stable, and the performance degradation of engine components is slow and negligible. Taking the most volatile ΔEGT as an example, arbitrarily selecting a set of data (14 points) from the entire lifespan operating data... Figure 4 As shown in the subplot, ΔEGT did not show any significant trend change during the 14 flight missions from August 21 to 27, 2009, and the data approximately followed the requirement of independent and identical distribution.

[0086] Therefore, the gas path parameter deviation values ​​are grouped using equal-length overlapping sliding windows. The Grubbs criterion is then used to remove gross errors from the grouped gas path parameter deviation values ​​following these steps:

[0087] (1) For a set of corrected gas path parameter deviations that follow a Gaussian distribution The mean μ and standard deviation σ of the data set can be calculated using equations (2) and (3).

[0088]

[0089] (2) This set of data Arrange in ascending order according to formula (4): d (1) <d (2) <…<d (n) (4),

[0090] (3) Calculate the sorted data according to formula (5). The corresponding statistic g (i) :

[0091] If data point d (i) The corresponding statistic g (i) >g (0) If (n, α), then this point represents a gross error and needs to be removed from the data set. Where g... (0) (n,α) is the critical value of the statistic when the number of repeated measurements is n and the confidence probability is α. It can be obtained by consulting the Grubbs critical value table.

[0092] (4) To ensure data integrity, the data d to be removed i The data can be supplemented using interpolation method (6).

[0093]

[0094] EEMD-SG-based filtering and noise reduction: This example uses EEMD-SG to filter and reduce the noise of the gas path parameter deviation value D after coarse error removal. The specific steps are as follows:

[0095] 1. Based on the deviation values ​​of gas path parameters decomposed by EEMD, EEMD solves the mode mixing problem faced by EMD by repeatedly superimposing Gaussian white noise into the original signal and utilizing the statistical characteristic that the mean of Gaussian white noise is 0. The specific steps are as follows:

[0096] (1) Set the number of repetitions m of the superimposed Gaussian white noise and the amplitude α of the white noise relative to the standard deviation of the original signal;

[0097] (2) According to formula (8), Gaussian white noise N with amplitude α is superimposed on the gas path parameter deviation value D after gross error removal. i The new signal obtained is denoted as

[0098] (3) Using EMD to analyze new signals The decomposition is performed, and it is decomposed into n intrinsic mode components I according to equation (9). i,j and one residual component R i ,

[0099] (4) Repeat steps (2) and (3) m times;

[0100] (5) The mean values ​​of each IMF component and residual component obtained by calculating according to equations (10) and (11) are used as the intrinsic mode components I after decomposition of the gas path parameter deviation value D. j and residual components R,

[0101]

[0102] Due to the decomposition of each intrinsic mode component I j It has the following two characteristics: 1) The number of extreme points and zero-crossing points is the same or differs by no more than 1; 2) The upper and lower envelopes are symmetrical along the time axis. It can be seen that, as... Figure 5 As shown, the high-frequency intrinsic mode components I of the gas path parameter deviation value after EEMD decomposition h The signal has a short period, contains few original signal trend components, is basically symmetrical along the time axis, and the component mean tends to 0; low-frequency intrinsic mode components I l The residual component R signal has a large period, contains a large number of trend components from the original signal, has a large deviation along the time axis, and the component mean is not 0.

[0103] Therefore, it can be considered that data noise is mainly contained in the high-frequency intrinsic mode components I. h In the above, the noise component is considered, while the signal trend is mainly reflected in the low-frequency intrinsic mode component I. l In the residual component R, it can be regarded as an evolutionary component, so data denoising mainly depends on the high-frequency intrinsic mode component I. h The processing.

[0104] Based on the characteristics of IMFs, a one-sample t-test can be used to perform a zero-mean test on the cumulative sum of IMF components to distinguish between high-frequency and low-frequency IMFs. Those that pass the test are high-frequency IMFs, and those that fail are low-frequency IMFs. The specific steps are as follows:

[0105] (1) Calculate the cumulative IMF component I according to formula (12). s ,

[0106] (2) Establish the test hypothesis and determine the test level.

[0107] Let the test mean be μ = 0, and the null hypothesis H0: μ Is =μ, Alternative hypothesis H1: μ Is ≠μ, significance level α=0.05, where μ Is For the cumulative IMF component I s The mean.

[0108] (3) Select the test method and calculate the test statistic. Use the one-sample T-test formula (13) to calculate the statistic and perform a two-tailed test.

[0109] In the formula, k——I s The number of samples;

[0110] σ Is —I s Standard deviation;

[0111] (4) Determine the P value and draw inferences, and calculate the t obtained according to equation (13). α,k-1 The value is determined by consulting the T-distribution table. If P > α, then H0 is accepted, and the IMF component I is accumulated. s Pass the zero-mean t-test; if P < α, then accept H1 and accumulate the IMF component I. s The zero-mean t-test was not passed when the cumulative IMF component that passed the test was I. d When j < d, the IMF components are high-frequency signals, and the IMF components with j > d are low-frequency signals.

[0112] Figure 6 The cumulative high-frequency IMF component I separated from a certain signal by a T-test h Cumulative low-frequency IMF component I l The residual component R and the original signal distribution show that the T test can effectively distinguish between the high-frequency and low-frequency IMF components decomposed by the EEMD method.

[0113] The SG filtering method is used to filter high-frequency IMF signals. The biggest advantage of this low-pass filter is its ability to reduce random noise while preserving data width and shape. For time-series data, the SG filter essentially performs a weighted average of the data within a window. The weighting coefficients are obtained by fitting a selected high-order polynomial. This is achieved by using an equal-length, overlapping sliding window along the time axis to fit the continuous time-series data within the window using a least-squares polynomial method, thereby obtaining the data distribution trend. The specific steps are as follows:

[0114] (1) For high-frequency IMF components I h Within an overlapping sliding window of equal length, there are 2m+1 consecutive values ​​x. i , i∈(-m,m), and fit the data by constructing a k-th order polynomial (k≤2m+1) according to equation (14).

[0115]

[0116] In the formula a j — Undetermined coefficients of a polynomial.

[0117] (2) Determine the undetermined coefficients a of the polynomial j The residual E between the data and the original data can be fitted using the least squares method according to equation (15).

[0118]

[0119] (3) To minimize the residual E, the reciprocal of E with respect to the undetermined coefficients of each polynomial should be 0.

[0120]

[0121] Solving for:

[0122]

[0123] The center point estimate of the window is calculated using the fitted polynomial, and then the high-frequency intrinsic mode components I at any time step are traversed using an equal-length overlapping sliding window. h The signal smoothing process can be completed to obtain the reconstructed high-frequency IMF signal I. h ′.

[0124] The reconstructed gas path parameter deviation value: According to equation (18), the high-frequency IMF component I obtained after smoothing by the SG filter is... h Low-frequency IMF component I l By summing the residual component R, we can obtain the reconstructed gas path parameter deviation value D′:

[0125]

[0126] In the formula I i — High-frequency IMF components processed by the SG filter;

[0127] I i —Low-frequency IMF components not processed by the SG filter;

[0128] R is the residual component obtained from EEMD decomposition.

[0129] The data preprocessing method proposed in this example is compared with existing conventional methods to verify the performance of the present invention:

[0130] For pure signal X ps and noisy signal X ns The following two metrics are typically used as evaluation criteria for data denoising effectiveness:

[0131] (1) Signal-to-noise ratio (SNR)

[0132] The signal-to-noise ratio is the ratio of the power of the clean signal to the power of the noise signal, and its calculation formula is as shown in equations (19)-(21).

[0133]

[0134]

[0135] In the formula P s ——Pure Signal X ps ; signal power;

[0136] x ij —Noisy signal X ns The signal power.

[0137] The higher the signal-to-noise ratio, the better the noise reduction effect, meaning that the data after noise reduction contains less noise.

[0138] (2) Root Mean Squared Error (RMSE)

[0139] The root mean square error is the pure signal X ps and noisy signal X ns The error between them is calculated using the formula (22).

[0140]

[0141] The smaller the root mean square error, the better the signal X contains noise. ns With pure signal X ps The smaller the error, the better the noise reduction effect.

[0142] Since the air path parameter deviations included in the collected engine operating data are all noisy signals and their pure signals cannot be obtained, the above evaluation indicators are difficult to directly apply to the evaluation of the noise reduction effect of air path parameter deviations. However, in order to demonstrate the effectiveness of the method studied in this example, without loss of generality, a randomized sample containing a DC component Y is constructed. dcs With the communication component Y acs The analog signal Y ps For comparing the noise reduction effects of different noise reduction methods, analog signal Yps The calculation formula is shown in equation (23).

[0143] Y acs =2sin(0.3x)cos(0.7x)+sin(0.1x+0.9)

[0144] Y dcs =0.5x + 4

[0145] Y ps =Y acs +Y dcs (twenty three)

[0146] In the formula, x is the input signal, and x ∈ (-2π, 2π).

[0147] Within the range of values ​​for the input signal x, 300 points are uniformly selected as the input signal. According to equation (19), the pure signal Y is then... ps Add 10dB of Gaussian noise and modify the values ​​of the input signal at points 50, 100, 150, 200, and 250 to -4, 10.5, -2, 11, and -1 to simulate gross errors. The analog signal Y generated by the above operations... ps Its noisy signal Y ns Signals such as Figure 7 As shown.

[0148] The simulated signal Y was analyzed using three experimental methods: the inertial filtering algorithm, the EMD-SVD filtering algorithm, and the EEMD-SG filtering algorithm, all employed in GE's diagnostic training manual. ps and noisy signal Y ns Noise reduction processing was performed. The noise-reduced signals after processing by each experimental method are shown below. Figure 2-8 As shown in Table 2, the corresponding evaluation indicators are as follows.

[0149] As shown in Table 2, for noise reduction filtering of noisy signals, the EEMD-SG filtering algorithm used in this example has a significant improvement in signal-to-noise ratio and root mean square error compared to the inertial filtering algorithm and EMD-SVD filtering algorithm used by GE.

[0150] Table 2 Evaluation Indicators for Each Noise Reduction Method

[0151]

[0152] Application examples of preprocessing gas path parameter deviation values

[0153] After verifying the effectiveness of the preprocessing method for gas path parameter deviation values ​​used in this invention through the aforementioned experiments, the gas path parameter deviation values ​​of the selected engine are still used as an example to perform data preprocessing step by step. Since the preprocessing steps for ΔEGT, ΔFF, and ΔN2 are completely identical, due to space limitations, the data preprocessing will only be demonstrated using ΔEGT as an example to illustrate the preprocessing process for gas path parameter deviation values.

[0154] First, using 14 points as a data set, select g (0) (14, 0.05) = 2.755. The Grubbs criterion is used to remove gross errors from the gas path parameter deviations. The gas path parameter deviations after gross error removal are as follows: Figure 9 As shown. Then, the number of repetitions of the superimposed Gaussian white noise was set to m = 100, and the amplitude of the white noise relative to the standard deviation of the original signal was set to α = 0.05. The gas path parameter deviation values ​​were decomposed using EEMD, and the decomposed IMF components are shown below. Figure 10 As shown.

[0155] Next, the cumulative IMF components obtained from the deviation values ​​of the gas path parameters decomposed by EEMD were determined using a single-sample zero-mean t-test to distinguish between high-frequency and low-frequency IMF components. The t-test results for each cumulative IMF component are shown in Table 2-3.

[0156] Table 3. Statistical table of T-test for the cumulative IMF components separated by ΔEGT.

[0157]

[0158]

[0159] Secondly, using 61 points as the width of an equal-length overlapping sliding window (m = 30), a 5th-order polynomial is constructed, and a Savitzky-Golay filter is used to smooth the high-frequency IMF signal I. h The IMF component of ΔEGT after filtering is as follows: Figure 11 As shown.

[0160] Finally, the gas path parameter deviation values ​​are reconstructed by combining the evolutionary components and the filtered noisy components. The gas path parameter deviation values ​​before and after reconstruction are as follows: Figure 12 As shown.

[0161] It can be seen that by using the Grubbs criterion and EEMD-SG to preprocess the deviation values ​​of the air path parameters, the influence of the engine's unsteady operating conditions on the deviation values ​​of the air path parameters is effectively eliminated, gross errors in the data are removed, and data noise is reduced.

[0162] To address the need for performance evaluation of air path components in civil aviation engines in existing technologies, this invention preprocesses air path parameter deviation values. First, fingerprint images are used to correct deviations in air path parameters caused by non-component performance degradation. Then, the Grubbs criterion is employed to remove gross errors from air path parameter deviation values ​​that meet the independent and identically distributed requirements within a short period, divided by equal-length overlapping sliding windows. Next, to address issues such as modal coupling and insufficient IMF signals based on correlation coefficients encountered with conventional noise reduction methods like EMD decomposition, EEMD decomposition is introduced. A single-sample T-test is used to distinguish between high- and low-frequency IMF signals. By applying SG filtering noise reduction only to high-frequency IMFs, the problem of long-term trend signal destruction due to data filtering noise reduction is avoided. Finally, comparative experiments verify the superiority of the proposed data preprocessing method in air path parameter deviation value preprocessing.

Claims

1. A method for preprocessing deviation values ​​of aero-engine air path parameters, characterized in that, Includes the following steps: Step 1: For the deviation value of the gas path parameters The correction formula based on the fingerprint image is as follows: （1）， In the formula, for i The original gas path parameter deviation value was not corrected at time 1; For the first i At time 1, the unit opening angle or unit bleed air leakage of VSV and VBV; For fingerprint image The offset caused by deviations in gas path parameters; Step 2: Use the Grubbs criterion to remove gross errors from the grouped gas path parameter deviations; Step 3: Use EEMD-SG to analyze the deviation values ​​of the gas path parameters after gross error removal. Noise reduction filtering is performed using the following steps: Step 3-1: Decompose the gas path parameter deviation values ​​based on EEMD; Step 3-2: Use a one-sample t-test to perform a zero-mean test on the cumulative sum of IMF components to distinguish between high-frequency and low-frequency IMFs. Those that pass the test are high-frequency IMFs, and those that fail are low-frequency IMFs. The specific steps are as follows: Step 3-2-1: Calculate the cumulative IMF component according to formula (12) I s : （12）， Step 3-2-2: Establish the test hypothesis and determine the significance level: Set the test mean. μ 0=0, null hypothesis H0: Alternative hypothesis H1: Significance level ,in To accumulate IMF components I s The mean; Step 3-2-3: Select the test method and calculate the test statistic: Calculate the statistic using the one-sample T-test formula (13) and perform a two-tailed test: （13）， In the formula, k 1 is I s Number of data samples; for I s Standard deviation; Step 3-2-4: Confirm P The value and the conclusion drawn: calculated according to formula (13) The value is obtained by consulting the T-distribution table to find the P-value. P > α 1. Accept H0 and accumulate IMF components. Pass the zero-mean t-test; like P<α If H1 is accepted, the IMF component will be accumulated. I s The zero-mean t-test was not passed. When the cumulative IMF component that passes the test is When the index in the IMF component satisfies j 2< d The IMF components are high-frequency IMF components, and their serial numbers satisfy... d < j The IMF component of 2 is a low-frequency IMF component; Step 4: Smooth the high-frequency IMF signal using a Savitzky-Golay filter. The specific steps are as follows: Step 4-1: For high-frequency IMF components The corresponding 2 within an inner equal-length overlapping sliding window m 2+1 consecutive values , Construct a by following formula (14) k A second-order polynomial is used to fit the data set, where k 2≤2 m 2+1: （14）， In the formula, The coefficients are undetermined polynomials. Step 4-2: Determine the undetermined coefficients of the polynomial The residuals between the fitted data and the original data are obtained by least squares method according to equation (15). : （15）， Step 4-3: To minimize the residual E, let E For the undetermined coefficients of each polynomial The derivative is 0: （16）， Solving for: （17）， The center point estimate of the window is calculated using the fitted polynomial, and then the high-frequency intrinsic mode components at any time step are traversed using an equal-length overlapping sliding window. Signal smoothing can then be completed to obtain the reconstructed high-frequency IMF signal. ; Step 5: Reconstruct the deviation value of the gas path parameters after noise reduction: According to formula (18), the high-frequency IMF component obtained after smoothing by the SG filter is... Low-frequency IMF components not processed by SG filter The residual components obtained from EEMD decomposition By summing the values, we can obtain the deviation values ​​of the reconstructed gas path parameters. , （18）。 2. The method for preprocessing deviation values ​​of aero-engine air path parameters according to claim 1, characterized in that, Step 2 is detailed below: Step 2-1: For a set of corrected gas path parameter deviations that follow a Gaussian distribution Calculate the mean of the data set using equations (2) and (3). with standard deviation , （2）， （3）， Step 2-2: Sort the data in ascending order according to formula (4): （4）； Step 2-3: Calculate the statistic corresponding to the sorted data group according to formula (5). : （5）， If data points Corresponding statistics Then the data point This is a gross error and needs to be removed from the data set. For repeated measurements of n², with a confidence probability of... α The critical value of the statistic at time 2 was obtained by consulting the Grubbs critical value table. Steps 2-4: To ensure data integrity, the data that was removed... The data is completed using the interpolation method of equation (6): （6）。 3. The method for preprocessing deviation values ​​of aero-engine gas path parameters according to claim 1, characterized in that, Step 3-1 specifically includes the following steps: Step 3-1-1: Set the number of repetitions for the superimposed Gaussian white noise. m 1 and the amplitude of white noise relative to the standard deviation of the original signal α 3, Step 3-1-2: According to formula (8), the deviation value of the gas path parameters after gross error removal. Mid-stack amplitude α 3 Gaussian white noise The new signal obtained is denoted as , （8）； Step 3-1-3: Use EMD to analyze the new signal Decompose it according to equation (9) into n 3 intrinsic mode components and 1 residual component : （9）， Step 3-1-4: Repeat m Steps 3-1-2 and 3-1-3 are performed once; The mean values ​​of each IMF component and residual component obtained by calculating according to formulas (10) and (11) are used as the deviation values ​​of the gas path parameters. D Intrinsic mode components obtained after EEMD decomposition and residual components : （10）， （11）， Data noise is contained in high-frequency intrinsic mode components. In the middle, it is considered as a noisy component; the signal trend is reflected in the low-frequency intrinsic mode components. In the residual component R, it is considered as an evolutionary component, so data denoising depends on the high-frequency intrinsic mode components. The process is then implemented.

Images