Differential value regression method for aero-engine outfield use performance calibration

Abstract

The invention provides an aero-engine complete machine performance calibration method based on difference regression. The aero-engine complete machine performance calibration method comprises four parts of parameter difference calculation at adjacent acquisition moments, clustering based on a Gaussian mixture model, model construction based on support vector regression and performance parameter correction. According to the method, the degradation amount of data collected at adjacent time points is eliminated by utilizing a difference value method, meanwhile, the data is divided into a plurality of categories by adopting a Gaussian mixture model, and the data of each category is fitted by adopting a support vector regression method, so that a more accurate regression model is obtained. Compared with other engine complete machine performance calibration methods, the difference regression method can be not influenced by performance degradation when a regression model is constructed, and especially for continuous working condition changes in aero-engine outfield work, a function relation expression of influences of different environment parameters and state parameters on engine performance parameters is effectively constructed; therefore, performance calibration of the engine is realized.

Description

Technical Field

[0001] The invention belongs to the field of aircraft engine whole machine performance degradation analysis, and solves the performance calibration problem of aircraft engines during field use by designing a difference regression method. Background Art

[0002] As the main power system of aircraft, the reliability of aircraft engines is crucial to flight safety. Aircraft engines operate under high temperature and high pressure for a long time and need to be constantly switched to meet the power requirements of different tasks. This leads to the continuous accumulation and degradation of key components of aircraft engines. When the degradation of aircraft engine components accumulates to a certain extent, their performance will deviate from normal levels or even fail. With the advancement of intelligent sensing and signal processing technology, predictive maintenance has become an important aspect of intelligent manufacturing. By analyzing and predicting the performance degradation of aircraft engines, more effective maintenance plans can be formulated to improve the overall performance of aircraft engines and extend the service life of aircraft engines.

[0003] In fact, steady-state data used for performance analysis are usually collected from multiple continuously changing working conditions. Affected by different operating conditions, the directly collected performance parameters cannot reflect the performance change trend of the engine. Therefore, it is necessary to unify the sensor signals collected under different working conditions to the same benchmark working condition to achieve the performance calibration of the engine. The Kmeans method is widely used to eliminate the influence of working condition changes on performance parameters, but it needs to assume the number of working conditions in advance, and after correction according to the working condition, the dimensionless data loses its original physical meaning. Compared with obtaining the dimensionless change trend, in the comprehensive performance monitoring of the engine, it is more desirable to unify the performance parameters under the same working condition to intuitively evaluate the engine performance, such as the exhaust temperature margin EGTM. The classic similarity conversion method can convert all steady-state data to the same working condition. However, the widely used standard similarity conversion formula needs to be modified in the actual application process to adapt to different engines. The modification of the standard similarity conversion formula usually requires experimental data support, and it is difficult to obtain a large amount of experimental data of the real engine. For this situation, another solution is to modify the data by constructing an engine mechanism model to achieve working condition unification. The mechanism model relies on complex thermodynamic calculations, requires in-depth exploration of the complex control mechanism of the engine, and also requires the model to be modified according to actual flight data to achieve accurate construction of the model. This method has a complex process and poor adaptability. One of the focuses of the present invention is to develop a simple operating procedure to unify various operating conditions under the same reference working condition to achieve performance calibration of the engine during field use. Summary of the invention

[0004] In order to solve the problem that the current aircraft engine field steady-state data collected from multiple working conditions is difficult to perform performance calibration, the present invention proposes a differential regression method (G-DRM) for aircraft engine field performance calibration, which can effectively solve the problem of aircraft engine whole machine performance calibration.

[0005] The difference regression method (G-DRM) for field performance calibration of aircraft engines includes four parts: parameter difference calculation at the time of phase acquisition, clustering based on Gaussian mixture model, support vector regression model construction and performance parameter correction. It uses the difference method to eliminate the degradation of the data collected at adjacent time points. At the same time, the Gaussian mixture model is used to divide the data into multiple categories, and the support vector regression method is used to fit the data of each category to obtain a more accurate regression model. Compared with other engine performance calibration methods, the difference regression method can be unaffected by performance degradation when constructing the regression model, especially for the continuous working condition changes in the field work of aircraft engines, and effectively constructs the functional relationship between different environmental parameters and state parameters on the engine performance parameters.

[0006] The present invention proposes a differential regression method for calibrating the performance of an aircraft engine in field use, comprising the following steps:

[0007] Step 1: Split the steady-state data set collected during the flight of the aircraft engine into environmental parameters and state parameters; the operating conditions of the aircraft engine are uniquely determined by the environmental parameters and state parameters;

[0008] Necessary notes for step 1: The steady-state data set of an aircraft engine includes three parts: environmental parameters, state parameters, and performance parameters; The steady-state data set of an aircraft engine The dataset contains I engine entities, and the i-th engine has K i steady-state data points, where t ik It represents the effective working time from the kth steady-state data sampling point of the i-th engine to the first sampling point of the engine. Represents the environmental parameters corresponding to the kth steady-state data point of the i-th engine, such as total atmospheric temperature, absolute pressure altitude, etc. Represents the flight state parameters corresponding to the kth steady-state data point of the i-th engine, such as throttle angle, true airspeed, speed, etc. Indicates the performance parameters corresponding to the kth steady-state data point of the i-th engine, such as exhaust temperature, output torque, etc. The aircraft engine environmental parameters include: absolute pressure altitude H a , relative pressure altitude H r 、Total atmospheric temperature T t 、Atmospheric static temperature T s; State parameters include: Mach number H, speed Ng, true angle of attack θ, true airspeed V t , indicated airspeed V i ;Performance parameters include: throttle angle PLA, torque τ, instantaneous consumption SFC.

[0009] Step 2, modeling the performance parameters and calculating the difference of the same parameter at adjacent collection points of the same engine;

[0010] The method includes a function f(E, S) of the operating condition determined by the performance parameter P, the environmental parameter E and the state parameter S, a function d(t, z) of the performance parameter decay value related to the effective flight working time t and the specific index z of each engine, and data acquisition noise ε, and the performance parameter is modeled as follows:

[0011] P = f(E, S) + d(t, z) + ε;

[0012] According to the maintenance information, the steady-state data of each engine is divided into multiple segments, and the difference between any two adjacent sampling points in each segment of each engine entity is calculated, including the difference ΔE of environmental parameters, ΔS of flight state parameters, and ΔP of performance parameters;

[0013] Necessary notes for step 2: The performance parameter modeling consists of the flight condition f(E,S), the performance degradation d(t,z) and the data acquisition noise. We assume that for the same aircraft engine, without maintenance, the performance change between any two adjacent acquisition points can be ignored, that is, d(t ik ,z i )=d(t i(k-1) ,z i ), but d(t i(k+1) ,z i )≠d(t i(k-1) ,z i ), and the difference of environmental parameters between any two adjacent collection points is calculated:

[0014]

[0015] Status parameter difference:

[0016]

[0017] Performance parameter difference:

[0018]

[0019] No performance degradation is involved.

[0020] Step 3: During the training phase, a difference regression model (G-DRM) combined with a Gaussian mixture model (GMM) is fitted.

[0021] Necessary notes for step 3: Before building the difference regression model, GMM is used to cluster the difference data to improve the accuracy of the model. As an extension of a single Gaussian probability density function, GMM can smoothly approximate density distributions of any shape and can use multiple distribution combinations to describe the difference steady-state data. Define a multidimensional random vector X = (ΔE, ΔS) T , the joint probability density function is expressed as the weighted sum of M multivariate Gaussian distribution functions:

[0022]

[0023] in is the parameter of the model. m ∈(0,1) is the weight of the mth Gaussian distribution, and the sum of all weights is 1, that is, N(x n |μ m ,∑ m ) is a multivariate Gaussian distribution with a mean of μ m , the covariance matrix is ​​Σ m In order to find the best M Gaussian distributions, the given data is used to estimate the parameters θ of the model. The maximum likelihood method is used to solve θ. Specifically, the EM algorithm consists of two steps: the first step is to calculate the probability that the nth sample belongs to the mth Gaussian distribution:

[0024]

[0025] The second step is to update the distribution parameters using the calculated probabilities:

[0026]

[0027] Repeat the above two steps until the model converges. Use the Gaussian distribution with the highest probability for each sample as the label of the sample. Use the negative log-likelihood function to select the model, which is the negative value of the following formula:

[0028]

[0029] The negative log-likelihood function is used to measure the explanation of the current model parameters for the known sample set. The smaller the value, the better, and the data is divided into the best m categories. The difference data environmental parameter difference ΔE, flight state parameter ΔS and performance parameter ΔP are clustered into m categories using GMM, and SVR is used to fit m categories to obtain m difference regression models. The regression model is ΔP=h(ΔE,ΔS).

[0030] Step 4: During the test phase, a reference operating point is selected, and the parameter differences between different acquisition points and the reference point of the same engine are calculated respectively. The difference regression model is input, and the original performance parameters are subtracted from the results to obtain the performance parameters after the unified operating conditions.

[0031] Necessary notes for step 4: The environmental parameters and state parameters of each engine's initial state are not exactly the same. Calculate the difference between all data and the standard working conditions, the environmental parameter difference:

[0032]

[0033] Status parameter difference:

[0034]

[0035] Performance parameter difference:

[0036]

[0037] ΔP0 is not only affected by ΔE0 and ΔS0, but also by the engine degradation effect d, that is,

[0038] ΔP0=h(ΔE0,ΔS0)+d(t,z)+ε

[0039] The GMM model learned in the training phase is combined with ΔE0 and ΔS0 to divide the samples into different categories. The SVR model corresponding to the category is used to predict the performance parameter change AP′0 that is not affected by the degradation effect, that is,

[0040] ΔP′0=h′(ΔE0,ΔS0)

[0041] Where b is the fitting deviation. Finally, the standard steady-state performance parameter P′ is calculated using the following formula:

[0042] P′=P0+(ΔP0-ΔP′0)=P-ΔP′0=P0+d(t,z)+ε+b

[0043] Where b=h(ΔE0, ΔS0)-h′(ΔE0, ΔS0) is the fitting error of the model.

[0044] Beneficial effects:

[0045] G-DRM eliminates the interference of performance degradation on performance parameter modeling by taking the difference between adjacent data points of the aircraft engine, and effectively constructs the functional relationship between the performance parameter difference and the environmental parameter difference and the state parameter difference, so as to better realize the performance calibration of the aircraft engine. At the same time, the Gaussian mixture model is used for clustering to better improve the fitting accuracy of the model. In addition, the performance parameters are restored according to the difference regression model so that all the collected data are unified under the same benchmark working conditions, which can show the performance degradation trend of the aircraft engine. It is applied to the performance degradation trend analysis of the aircraft engine as a whole to complete the performance calibration of the aircraft engine during field use. Through a large number of performance degradation trend extraction experiments on a certain type of turboprop engine field use data, it is shown that compared with traditional data-driven methods and many data conversion methods, G-DRM has higher accuracy in the field of aircraft engine performance calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is the distribution diagram of turboprop engine data;

[0047] Figure 2 It is the trend chart of raw data of turboprop engine;

[0048] Figure 3 It is the correlation coefficient diagram between the original data performance variables and the operating condition variables;

[0049] Figure 4 is the box plot of the operating condition variables;

[0050] Figure 5 It is a clustering result display diagram;

[0051] Figure 6 is a graph of degradation characteristics trends over time;

[0052] Figure 7 is the average Spearman correlation coefficient plot between degradation characteristics and operating condition variables;

[0053] Figure 8 This is the trend graph of EGT′ changing with time in TP007;

[0054] Fig. 9 This is the trend graph of EGT′ changing with time in TP008;

[0055] Fig.10 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0056] This case takes a certain type of airborne turboprop engine as the research object. This type of engine is widely used in large transport aircraft and some rescue aircraft. Specifically, this type of engine is a fixed turbine single-rotor turboprop engine, which uses an axial flow compressor to provide compressed air for the combustion chamber, and transmits the shaft power output by the engine to the propeller through the reducer. During each flight, multiple steady-state working data are collected from the flight parameters for the overall performance analysis of the engine. As shown in Table 1, we have counted the steady-state data of 12 engines during the entire field use cycle. The sensor parameters collected at each steady-state point are shown in Table 2, where EGT, SFC and τ are the performance parameters of this type of engine, that is, the changing trends of these parameters over time reflect the performance changes of the engine. Further plotting the changing trends of some parameters over time in the data collected from engines TP001, TP007, TP008 and TP012 yields the following results. Figure 1 It can be found that the working conditions of the steady-state data collected during the life cycle of the engine are constantly changing over time, and due to the influence of the changes in working conditions, it is impossible to directly judge the performance changes of the engine by observing the changing trends of performance parameters. Therefore, it is necessary to use certain methods to unify the engine steady-state data collected under various working conditions to the same benchmark working condition to achieve the calibration of engine performance.

[0057] Table 1. Turboprop engine steady-state data set

[0058]

[0059] Table 2 Summary of sensor measurement variables

[0060]

[0061] In order to verify the superiority of the proposed G-DRM model in extracting the degradation characteristics of aircraft engines, five comparative methods were used in the experiment. The K-means method classifies the steady-state data into K different categories based on the environmental variable E and the state variable S. Subsequently, the Z-score normalization is applied independently to each category, and the normalized data is sorted in chronological order. Finally, multiple experiments were carried out and the best results were recorded. The standard similarity conversion model (SSCM) is a performance parameter correction method widely used in the field of aircraft engines. This method corrects the steady-state data collected under various operating conditions to standard atmospheric conditions, effectively avoiding the influence of operating condition changes. For the turboprop engine studied, no specific similarity conversion formula is given in the manual. Therefore, a general standard similarity conversion model is used to correct the performance parameters. The SVR regression method directly uses the SVR fitting regression model P = f (E, S), and obtains the degradation amount by subtracting the result of the model fitting from the original performance parameter, that is, The G-SVR method is similar to the G-DRM method. Before building the regression model, the original data is classified using the GMM method, and then the SVR is applied to fit the regression model for each category. Furthermore, in order to illustrate the impact of GMM on the performance calibration results, we use the DRM method without GMM clustering to directly fit the difference model.

[0062] In order to evaluate the effectiveness of different methods in extracting degradation features, we defined three evaluation indicators: correlation change, monotonicity, and trend consistency. Correlation change indicator, the correlation coefficient is used to quantify the relationship between variables. The goal of extracting degradation features is to unify the data to the same benchmark operating conditions. It can be foreseen that after this process, the correlation between performance variables and operating condition variables will be significantly reduced. Therefore, the effectiveness of different degradation feature extraction methods can be evaluated by calculating the correlation change before and after degradation feature extraction. Unlike the Pearson correlation coefficient, the Spearman correlation coefficient is more suitable for nonlinear data relationships. Therefore, we use the Spearman correlation coefficient to measure the correlation between engine performance variables and operating condition variables. The calculation formula is as follows:

[0063]

[0064] Where n is all the collected steady-state data, X = (E, S) T In addition, based on the Spearman correlation coefficient, the correlation change index is defined as follows:

[0065]

[0066] In the calculation, only the operating parameters with a correlation greater than 0.1 with the performance parameters are considered. Δcorr = 0 means that the correlation between the operating parameters and the performance parameters increases after the degradation feature extraction. The monotonicity index is used to evaluate the monotonicity of the degradation feature. Specifically, we calculated the average Spearman correlation coefficient between the degradation feature of each engine and the operating time of the aircraft engine to indicate the monotonicity of its time-varying changes. The calculation expression is as follows:

[0067]

[0068] in, represents the Spearman correlation coefficient between the jth performance parameter and its operating time in the i-th aircraft engine. As time goes by, the trend of the same degradation characteristics should be consistent, so the following trend consistency index is defined:

[0069]

[0070] When Tr jWhen it is 0 or 1, it means that the degradation characteristic P' of all aircraft engines j There is a consistent trend over time. j = 0.5 represents the degradation characteristic P' of all engines j It has nothing to do with the running time. j Any other value of , which indicates a degenerate characteristic P' j There are differences in trends over time.

[0071] In actual scenarios, our goal is to select performance parameters that have consistent and obvious changing trends over time to calibrate engine performance. When evaluating the effects of different methods, the trend consistency index, namely Tr j The required value is 0 or 1. Next, focus on the monotonicity index, The larger the value, the more obvious the trend of the extracted degradation features. Finally, consider the correlation change index, where Δcorr(P i ) is larger, the more significant the decrease in correlation is.

[0072] The calculation results of Spearman correlation coefficients for aircraft engines TP001, TP007, TP008 and TP012 are as follows: Figure 3 As shown. Comparing the calculation results of the four engines, it can be found that the correlation between the same performance parameter and the same operating condition parameter is close. This shows that although there are specific indicators z between different engines, their basic characteristics are still similar, such as the significant positive correlation between the total atmospheric temperature and the exhaust temperature. Therefore, the average correlation coefficient of multiple engines is used in the subsequent analysis. Further analysis Figure 3 , it can be found that the correlation between the same performance parameters and the running time in the four engines is not consistent. Specifically, the EGT of TP001 and TP003 is negatively correlated with the running time, while the EGT of TP007 and TP012 is positively correlated with the running time. This is due to the dynamic switching of the operating conditions. Therefore, it is necessary to correct the performance parameters. In addition, it can be seen from Table 2 that for several different physical values, multiple variables of the same type are recorded. For example, the altitude parameter records the absolute pressure altitude H at the same time. a and relative pressure altitude H r In order to enhance the model's fitting ability, for the same type of operating parameters, the corresponding variable with the highest correlation with the performance variable is selected as the representative of this type. Therefore, the operating parameters used for subsequent analysis are PLA, H a , T t , M, Ng and θ.

[0073] Figure 4The following is a distribution diagram of the operating condition variables in the original data. From the figure, we can see that there are a large number of outliers in the data. In order to prevent outliers from affecting the fitting accuracy of the model, the Tukey's Test method is used to identify outliers. Specifically, in the Tukey's Test method, outliers are identified as observations outside the following thresholds:

[0074] Lower Bound=Q1-k(Q3-Q1)

[0075] Upper Bound=Q3+k(Q3-Q1)

[0076] Where Q1 represents the lower quartile, Q3 represents the upper quartile, and k is a coefficient. k = 3 means filtering extreme outliers, while a value of k = 1.5 means filtering moderate outliers. Figure 4 (b) shows the filtering result when k = 1.5. Figure 4 (c) shows the filtering result when k = 3. Figure 4 (b) and Figure 4 (c), the result of k = 1.5 was finally selected for subsequent analysis.

[0077] During the experiment, the parameter K of Kmeans was selected in the range of 0 to 30. Finally, the best result was obtained at K=28, and Table 3 only records the results when K=28. Figure 5 (a) shows the clustering results when K = 28, where different colors represent different clusters. For G-SVR and G-DRM, the maximum number of Gaussian components is set to 30. The number of clusters corresponding to the minimum negative log-likelihood function is selected as the optimal number of Gaussians, which are 28 and 24 for G-SVR and G-DRM, respectively. Figure 5 (b) shows the data partitioning results of G-SVR. Figure 5 (c) shows the data partitioning results of G-DRM. It is worth noting that G-SVR performs clustering based on (P, X), while G-DRM uses (ΔP, ΔX) for clustering. In SVR, G-SVR, DRM and G-DRM, radial basis function (RBF) is used as the kernel function, and the penalty parameter C, the width γ of the radial basis function kernel function and the relaxation parameter ε are determined by 5-fold cross validation for each cluster.

[0078] Table 3 Turboprop engine degradation characteristics analysis results

[0079]

[0080]

[0081] Table 3 lists the evaluation results of different models on the turboprop engine steady-state data set, where RMSE is the average calculation result of 5-fold cross validation. As shown in Table 3, only the trends of EGT' of all engines corrected by DRM and G-DRM are consistent, while the trends of SFC' and τ' corrected by all methods are inconsistent. Figure 6 The degradation feature extraction results of engines TP007 and TP008 are shown. Over time, the trends of SFC' and τ' are not obvious, while the trend of EGT' is monotonically increasing. Compared with ground tests, the systematic errors of SFC and τ measurements during flight are larger. Therefore, EGT is more suitable as a performance indicator for aircraft engines. Further analysis of Table 3 shows that the EGT' corrected by G-DRM has a higher correlation with time, that is, the monotonicity is stronger, and the change in the correlation coefficient before and after correction is more significant. In addition, compared with SVR and G-SVR, the DRM and G-DRM models show higher fitting accuracy, that is, smaller RMSE. Figure 7 The correlation between the degradation features extracted by different methods and the operating condition variables is shown, where the correlation is taken as the average value of multiple engines and the calculation formula is as follows:

[0082]

[0083] It can be found that the correlation between all performance parameters and operating parameters has decreased, but only the correlation between EGT' corrected by DRM and G-DRM and all operating variables is less than 0.2. However, it can be clearly seen from Table 3 that the Δcorr corresponding to EGT' extracted by SSCM is 0. This is due to the special single-rotor structure of the turboprop engine and the data is collected during flight. Therefore, SSCM cannot be directly used for performance parameter correction of this engine.

[0084] Figure 8 and Fig. 9 The trend of EGT' corrected by different methods over time is shown, where the dotted line indicates the time when maintenance activities are performed. It can be seen that the EGT' corrected by DRM and G-DRM shows significant performance recovery after maintenance, that is, EGT decreases. In addition, by comparing DRM with G-DRM and SVR with G-SVR, it can be seen that the fitting accuracy of the model can be improved by dividing the data into multiple Gaussian distributions before regression fitting, so that the corrected performance parameters show stronger monotonicity and more significant correlation changes. Therefore, we use the EGT' corrected by G-DRM as the basis for the performance calibration of this turboprop engine. Figure 6Further analysis shows that the EGT trends of TP007 and TP008 over time are different. Compared with TP007, TP008 has an additional initial performance degradation stage. Combined with engineering practice analysis, this stage is the performance running-in stage of the engine. Further review of the data statistics process of TP007 found that there were initial data statistical omissions in TP007. In summary, the turboprop engine is mainly divided into three performance change stages, namely the performance running-in period, the performance platform period and the performance decline period. Maintaining the engine during the performance degradation stage can effectively extend the service life of the engine.

Claims

1. A difference regression method for field performance calibration of aircraft engines, characterized in that: The steps include: Step 1, splitting the steady-state data set collected during the flight of the aircraft engine into environmental parameters and state parameters; the operating conditions of the aircraft engine are uniquely determined by the environmental parameters and the state parameters; Steady-state dataset Contains, engine entities, the i-th engine has collected a total of K i steady-state data points, where t ik It represents the effective working time from the kth steady-state data sampling point of the i-th engine to the first sampling point of the engine. represents the environmental parameters corresponding to the kth steady-state data point of the i-th engine, represents the flight state parameter corresponding to the kth steady-state data point of the i-th engine, represents the performance parameter corresponding to the kth steady-state data point of the i-th engine; Step 2, model the parameters and calculate the difference of the same parameter at adjacent collection points of the same engine; Specifically, it includes: a function f(E, S) that determines the working condition based on the performance parameter P, the environmental parameter E and the state parameter S, a function d(t, z) of the performance parameter decay value related to the effective flight working time t and the specific index z of each engine, and data acquisition noise ε, and the performance parameter modeling is P=f(E, S)+d(t, z)+ε; According to the maintenance information, the steady-state data of each engine is divided into multiple segments, and the difference between any two adjacent sampling points in each segment of each engine entity is calculated to obtain the difference ΔE of environmental parameters, the difference △S of flight state parameters, and the difference AP of performance parameters; Step 3: During the training phase, a difference regression model G-DRM combined with a Gaussian mixture model GMM is fitted; Specifically, before building the regression model, GMM is used to cluster the difference data environmental parameter difference ΔE, flight state parameter difference △S and performance parameter difference AP, and support vector regression SVR is performed according to different clusters. The regression model is ΔP = h (ΔE, ΔS); Step 4: During the test phase, a reference operating point is selected, and the parameter differences between different acquisition points and the reference point of the same engine are calculated respectively, and the difference regression model is input, and the original performance parameters are subtracted from the model output results to obtain the performance parameters after the unified operating conditions; Specifically, it includes: calculating the difference ΔE0 between the environmental parameters and the benchmark operating point, the difference ΔS0 between the state parameters and the benchmark operating point, and the difference ΔP0 between the data performance parameters and the benchmark operating point; dividing the samples into different classes according to the GMM model learned in the training stage combined with ΔE0 and ΔS0; using the SVR model corresponding to the class to predict the performance parameter change ΔP0 that is not affected by the degradation effect, that is, ΔP0 = h′(ΔE0, ΔS0), where b is the fitting deviation; finally, the original performance parameter P is subtracted from the model output result ΔP0 to obtain the performance parameter P′ after the unified operating condition, that is, P′ = P-ΔP0.

2. The difference regression method for field performance calibration of an aero-engine according to claim 1, characterized in that: In step 1, the aircraft engine environmental parameters include: absolute pressure altitude H a , relative pressure altitude H r 、Total atmospheric temperature T t 、Atmospheric static temperature T s etc.; state parameters include: throttle angle PLA, Mach number H, speed Ng, true angle of attack θ, true airspeed V t , indicated airspeed V i Etc.; performance parameters include: exhaust temperature EGT, torque τ, instantaneous consumption SFC.

3. The difference regression method for field performance calibration of an aero-engine according to claim 1, characterized in that: Step 2: For the same aircraft engine, without maintenance, the performance change of any two adjacent acquisition points is ignored, that is, d(t ik , z i )=d(t i(k-1) , z i ), but d(t i(k+1) , z i )≠d(t i(k-1) , z i ), and the difference of environmental parameters between any two adjacent collection points is calculated Flight status parameter difference Performance parameter difference No performance degradation is involved.

4. The difference regression method for field performance calibration of an aero-engine according to claim 1, characterized in that: In step 3, before constructing the difference regression model, GMM is used to cluster the difference data to improve the accuracy of the model; a multidimensional random vector X = (△E, △S) consisting of the environmental parameter difference ΔE and the flight state parameter difference ΔS is defined. T , the joint probability density function of the vector is expressed as the weighted sum of M multivariate Gaussian distribution functions: in is the parameter of the model, π m ∈(0, 1) is the weight of the mth Gaussian distribution, and the sum of all weights is 1, that is, N(x n |μ m ,∑ m ) is a multivariate Gaussian distribution with a mean of μ m , the covariance matrix is ​​∑m; in order to find the best M Gaussian distributions, the given data is used to estimate the parameters θ of the model; the maximum likelihood method is used to solve θ; specifically, the EM algorithm consists of two steps Composition: The first step is to calculate the probability that the nth sample belongs to the mth Gaussian distribution: The second step is to update the distribution parameters using the calculated probabilities: Repeat the above two steps until the model converges; use the Gaussian distribution with the highest probability for each sample as the label of the sample; use the negative log-likelihood function to select the model, which is the negative value of the following formula: The negative log-likelihood function is used to measure the explanation of the current model parameters for the known sample set. The smaller the value, the better. Based on this, the data is divided into the best m categories.

5. The difference regression method for field performance calibration of an aero-engine according to claim 1, characterized in that: In step 4, the initial installed state of each engine is used as the standard operating condition (E i0 , S i0 ), due to the influence of the specific index z, the performance parameters of each engine measured under the same state have deviations, that is, P i0 ≠P j0 .

6. The difference regression method for field performance calibration of an aero-engine according to claim 5, characterized in that: In step 4, the environmental parameters and state parameters of each engine in the initial state are not exactly the same. The difference between all data and the standard working conditions is calculated. State parameter difference Performance parameter difference 7. The difference regression method for field performance calibration of an aircraft engine according to claim 1, characterized in that: In step 4, △P0 is affected not only by △E0 and △S0, but also by the engine degradation effect d, that is, △P0=h(△E0, △S0)+d(t,z)+ε; the samples are divided into different classes according to the GMM model learned in the training phase combined with △E0 and △S0.

8. The difference regression method for field performance calibration of an aero-engine according to claim 1, characterized in that: In step 4, the standard steady-state performance parameter P′ is calculated using the following formula: P′=P0+(△P0-△P′0)=P-△P′0=P0+d(t,z)+ε+b, where P is the measured value of the performance parameter of the target data, P0 is the measured value of the benchmark condition, △P0=P-P0 is the difference between the measured values ​​of the performance parameter of the target data and the benchmark condition, △P′0=h′(△E0, △S0) is the model prediction value, and b=h(△E0, △S0)-h′(△E0, △S0) is the fitting error of the model.

Images