Method and device for self-correcting input parameters of aircraft engine onboard adaptive model

Abstract

The present invention discloses a method for self-correcting input parameters of an aircraft engine onboard adaptive model. The method of the present invention uses an additional auxiliary onboard model and a fan nonlinear speed controller to perform online correction on the measured values ​​of the input parameters of the aircraft engine onboard adaptive model, and updates the auxiliary onboard model by considering the performance degradation of engine components during the entire life cycle. The present invention also discloses a device for self-correcting input parameters of an aircraft engine onboard adaptive model and an aircraft engine control system. Compared with the prior art, the present invention can effectively improve the online correction accuracy of the input parameters of the aircraft engine onboard adaptive model during the entire life cycle to meet the requirements of engineering applications.

Description

Technical Field

[0001] The invention relates to a method for self-correcting input parameters of an aircraft engine onboard adaptive model, and belongs to the technical field of aircraft engine control. Background Art

[0002] The concept of aircraft engine model-based control was first proposed in 1992. Subsequently, major foreign aircraft engine companies have evaluated the feasibility of model-based performance control and the considerable performance benefits it brings through simulation tests. In the past 10 years, NASA has conducted in-depth research on the relevant technical issues of airborne adaptive models in engine model-based performance control, and has successively proposed linear / nonlinear airborne adaptive models based on constant gain extended Kalman filters and reduced dimension Kalman filters based on singular value decomposition / optimal adjustment to estimate engine performance parameters. Its adaptive ability enables model-based performance control to cover the entire life cycle of the engine, and bench tests and airborne flight experiments have been realized based on the developed CMAPSS40K simulation software. In China, various aviation universities and related research institutes have also successively carried out research on aircraft engine model-based performance control methods, but the main technical means and principle architecture are referenced from NASA in the United States. Due to the fact that key technical issues such as its real-time performance and accuracy have not yet been resolved, it is still in the stage of theoretical research and digital simulation, and it is difficult to reach the level of engineering application.

[0003] The existing airborne adaptive models are mainly divided into two types: one is a linear airborne adaptive model designed by combining a linear airborne model + a linear Kalman filter, which is characterized by high real-time performance but low accuracy; the other is a nonlinear airborne adaptive model designed by combining a nonlinear airborne model + a nonlinear Kalman filter, which is characterized by high accuracy but poor real-time performance.

[0004] As the digital twin of the real engine, the input parameter value of the airborne adaptive model must always be the same as that of the real engine. In practice, it is found that the input parameters of the engine such as the fuel quantity are mainly measured by sensors such as float level gauges, volume flow meters, and mass flow meters. The measurement noise and measurement deviation, as well as uncertainty factors such as fuel freezing in extreme environments, often cause low measurement accuracy and insufficient reliability. At this time, the measured value of the sensor is not necessarily the real input parameter value of the engine. If it is directly used as the input of the airborne adaptive model, the airborne adaptive model will not be able to truly match the working state of the engine, resulting in insufficient parameter estimation. Although the current method (Song Zhiping et al., A correction method for fuel flow of a gas turbine engine, CN202010235155.9) can solve the problem of input parameter measurement uncertainty by matching the low-pressure shaft speed to estimate the fuel correction amount, it only corrects the fuel quantity for the dynamic physical model, that is, by constructing a dynamic physical model of the gas turbine engine, the fuel correction amount is obtained according to the deviation between the dynamic physical model and the low-pressure shaft speed of the real engine, so as to match the fuel measurement value. Subsequently, in order to improve the correction accuracy, the literature (Liu Jinxin et al., Gas Turbine Engine Fuel Flow Correction Method Based on Network Model, CN202010241059.5) trained a fuel correction network model through data, and made secondary corrections to the fuel correction amount until the dynamic physical model matched the actual engine speed to obtain a high-precision fuel correction amount. However, current studies have not considered the impact of actual conditions such as engine component performance degradation during the entire life cycle on fuel flow correction, and the application is still difficult.

[0005] Therefore, in practice, problems such as real-time performance and uncertainty in input parameter measurement of real engines are still the main technical challenges that seriously restrict the engineering application of airborne adaptive models. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the uncertainty problem of input parameter measurement of existing aircraft engine onboard adaptive model in engineering applications, and to provide an aircraft engine onboard adaptive model input parameter self-correction method, which can effectively improve the online correction accuracy of aircraft engine onboard adaptive model input parameters throughout the life cycle to meet engineering application requirements.

[0007] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0008] A method for self-correcting input parameters of an aircraft engine onboard adaptive model, wherein the aircraft engine onboard adaptive model comprises a main onboard model and a Kalman filter, and is used for online estimation of aircraft engine performance parameters including health parameters; an additional auxiliary onboard model and a fan nonlinear speed controller are used to perform the following corrections on the input parameter measurement values ​​of the aircraft engine onboard adaptive model: the fan speed measured in real time by the real engine is used as a reference value, the fan speed estimated by the auxiliary onboard model is used as a feedback value, and the fan nonlinear speed controller is used to correct the fan nonlinear speed of the aircraft engine onboard adaptive model by taking the error between the fan speed reference value and the feedback value as input and taking the input parameter measurement value as a feedforward value. The speed controller calculates the input parameter measurement value deviation, and corrects the input parameter measurement value with the calculated input parameter measurement value deviation and then inputs it into the aircraft engine onboard adaptive model; and when the real aircraft engine is in a steady state and the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is greater than a preset error threshold, the health parameter degradation amount of the real engine estimated by the aircraft engine onboard adaptive model is filtered, and then the auxiliary airborne model is updated with the filtered health parameter degradation amount until the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is less than or equal to the error threshold.

[0009] Based on the same inventive concept, the following technical solutions can also be obtained:

[0010] A device for self-correcting input parameters of an aircraft engine onboard adaptive model, the aircraft engine onboard adaptive model comprising a main onboard model and a Kalman filter, for online estimation of aircraft engine performance parameters including health parameters; the device comprising an additional auxiliary onboard model, a fan nonlinear speed controller and a health parameter update module; the auxiliary onboard model and the fan nonlinear speed controller are used to perform the following corrections on the input parameter measurement values ​​of the aircraft engine onboard adaptive model: using the fan speed measured in real time by the real engine as a reference value, using the fan speed estimated by the auxiliary onboard model as a feedback value, and using the error between the fan speed reference value and the feedback value as input and using the input parameter as the input parameter. The fan nonlinear speed controller using the numerical measurement value as the feedforward quantity calculates the input parameter measurement value deviation, and corrects the input parameter measurement value with the calculated input parameter measurement value deviation and then inputs it into the aircraft engine airborne adaptive model; the health parameter updating module is used to filter the health parameter degradation amount of the real engine estimated by the aircraft engine airborne adaptive model when the real aircraft engine is in a steady state and the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is greater than a preset error threshold, and then use the filtered health parameter degradation amount to update the auxiliary airborne model until the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is less than or equal to the error threshold.

[0011] Preferably, the filtering process is performed using the following outlier-median average filtering algorithm:

[0012] S1. Obtain the health parameter data set D(t,h) estimated by the onboard adaptive model at the N moments before the current moment;

[0013] S2, for each data point t in the data set D(t,h) i , set an internal window and set t i Compared with all other data points in this window j Perform triple cubic weight function calculation to obtain weight ζ ij :

[0014]

[0015] Wherein, d is the width of the internal window;

[0016] S3. In the internal window of each data point, a polynomial is fitted using the weighted least squares method. The goal is to minimize the weighted residual sum of squares to obtain a new data set Data wnd :

[0017]

[0018] Among them, p(t i ; β) is a polynomial model, β is the coefficient of the polynomial, h j is the jth data of health parameter;

[0019] S4. Discard Data wnd The minimum and maximum values ​​in the filter are averaged to obtain the degradation amount of the health parameter at the current moment after filtering.

[0020]

[0021] Preferably, the Kalman filter is an improved spherical unscented Kalman filter estimator based on minimum sigma points, and the working process is as follows:

[0022] Step 1, k = 0, initialize the posterior state estimate and the posterior state estimation error covariance matrix

[0023]

[0024] Step 2: Perform singular value decomposition SVD:

[0025]

[0026] Among them, U k , S k , They represent the k-time pairs Perform SVD to obtain the unitary matrix, singular value matrix, and transposed matrix of the unitary matrix;

[0027] Step 3: Calculate the weighted coefficient W of the mean and covariance matrix:

[0028]

[0029] Among them, W0, W1, …, W n , are the weight coefficients corresponding to the 1st, 2nd, …, n+1th sigma points respectively, and the parameter vector δ is a non-zero adjustment parameter;

[0030] Step 4: Estimate the posterior state at time k Calculate n+1 sigma points:

[0031]

[0032] in, I is the identity matrix;

[0033] Step 5: Time update:

[0034] According to the known discrete state equation of the aircraft engine nonlinear system f(·,·), n+1 sigma points σ aug,i,k Convert to

[0035]

[0036] σ aug,i,k represents the i-th sigma point at time k, i is the index of the sigma point, i = 0, 1, ..., n, n is the dimension of the state quantity x; u k+1 represents the control input at time k+1; Indicates that according to the i-th sigma point σ aug,i,k and u k+1 The calculated posterior state sample value at time k+1;

[0037] Then the prior state estimate at time k+1 is for:

[0038]

[0039] W i Represents the weight coefficients of the state estimation mean and error covariance matrix;

[0040] Next, we get the prior state estimation error covariance matrix at time k+1: for:

[0041]

[0042] Q k represents the process noise matrix of the system at time k;

[0043] Step 6: Measurement update:

[0044] Step 1: Calculate the estimated output based on the known nonlinear system output function g(·,·)

[0045]

[0046] Prior output at time k+1 estimate:

[0047]

[0048] Step 2: Considering the measurement noise, the a priori output estimation error covariance matrix at time k+1

[0049]

[0050] The estimated error cross-covariance matrix between the prior state and the output at time k+1

[0051]

[0052] Step 7: Posterior state estimation and the posterior state estimation error covariance matrix renew:

[0053]

[0054] in, is the covariance matrix scaling factor;

[0055] Step 8, k=k+1, propagate and Return to step 2 and extract The degradation amount of health parameters in is used to update the host onboard model.

[0056] Further preferably, the nonlinear system output function g(·,·) is a nonlinear prediction model obtained by offline training using test data.

[0057] An aircraft engine control system includes an aircraft engine onboard adaptive model, wherein the aircraft engine onboard adaptive model includes a main onboard model and a Kalman filter, and is used for online estimation of aircraft engine performance parameters including health parameters; the aircraft engine control system also includes an aircraft engine onboard adaptive model input parameter self-correction device as described in any of the above technical solutions.

[0058] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0059] The present invention solves the influence of the measurement uncertainty of the real engine input parameters on the matching of the airborne adaptive model to the real engine state, and realizes the online self-correction of the input parameters of the airborne adaptive model; the present invention further uses the improved spherical unscented Kalman filter estimator to construct the airborne adaptive model, which effectively improves the calculation real-time and convergence of the airborne adaptive model. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of the structural principle of a preferred embodiment of an aircraft engine control system of the present invention;

[0061] Figure 2 for Figure 1 A schematic diagram of the self-correction process of input parameters of an aircraft engine onboard adaptive model of an aircraft engine control system is shown. DETAILED DESCRIPTION

[0062] In order to solve the problem of uncertainty in the measurement of aircraft engine input parameters, the present invention uses an additional auxiliary airborne model and a fan nonlinear speed controller to perform online correction on the input parameter measurement values ​​of the aircraft engine onboard adaptive model, and updates the auxiliary airborne model by considering the performance degradation of engine components throughout the life cycle, thereby eliminating the influence of the actual engine component performance degradation on the estimation accuracy of the input parameter correction value.

[0063] The present invention specifically adopts the following technical solutions to solve the above technical problems:

[0064] A method for self-correcting input parameters of an aircraft engine onboard adaptive model, wherein the aircraft engine onboard adaptive model comprises a main onboard model and a Kalman filter, and is used for online estimation of aircraft engine performance parameters including health parameters; an additional auxiliary onboard model and a fan nonlinear speed controller are used to perform the following corrections on the input parameter measurement values ​​of the aircraft engine onboard adaptive model: the fan speed measured in real time by the real engine is used as a reference value, the fan speed estimated by the auxiliary onboard model is used as a feedback value, and the fan nonlinear speed controller is used to correct the fan nonlinear speed of the aircraft engine onboard adaptive model by taking the error between the fan speed reference value and the feedback value as input and taking the input parameter measurement value as a feedforward value. The speed controller calculates the input parameter measurement value deviation, and corrects the input parameter measurement value with the calculated input parameter measurement value deviation and then inputs it into the aircraft engine onboard adaptive model; and when the real aircraft engine is in a steady state and the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is greater than a preset error threshold, the health parameter degradation amount of the real engine estimated by the aircraft engine onboard adaptive model is filtered, and then the auxiliary airborne model is updated with the filtered health parameter degradation amount until the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is less than or equal to the error threshold.

[0065] Based on the same inventive concept, the following technical solutions can also be obtained:

[0066] A device for self-correcting input parameters of an aircraft engine onboard adaptive model, the aircraft engine onboard adaptive model comprising a main onboard model and a Kalman filter, for online estimation of aircraft engine performance parameters including health parameters; the device comprising an additional auxiliary onboard model, a fan nonlinear speed controller and a health parameter update module; the auxiliary onboard model and the fan nonlinear speed controller are used to perform the following corrections on the input parameter measurement values ​​of the aircraft engine onboard adaptive model: using the fan speed measured in real time by the real engine as a reference value, using the fan speed estimated by the auxiliary onboard model as a feedback value, and using the error between the fan speed reference value and the feedback value as input and using the input parameter as the input parameter. The fan nonlinear speed controller using the numerical measurement value as the feedforward quantity calculates the input parameter measurement value deviation, and corrects the input parameter measurement value with the calculated input parameter measurement value deviation and then inputs it into the aircraft engine airborne adaptive model; the health parameter updating module is used to filter the health parameter degradation amount of the real engine estimated by the aircraft engine airborne adaptive model when the real aircraft engine is in a steady state and the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is greater than a preset error threshold, and then use the filtered health parameter degradation amount to update the auxiliary airborne model until the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is less than or equal to the error threshold.

[0067] In order to facilitate public understanding, the technical solution of the present invention is described in detail below through a preferred embodiment in conjunction with the accompanying drawings:

[0068] like Figure 1 As shown, the aircraft engine control system of this embodiment mainly includes the following three parts:

[0069] (1) The first onboard model as an auxiliary onboard model and the fan nonlinear speed controller:

[0070] The first onboard model and the fan nonlinear speed controller function to correct the measured input parameter values; e.g. Figure 1 As shown, the fan speed Nf is measured in real time by the real engine m As a reference value, the fan speed Nf estimated by the first onboard model e As the feedback value, the fan speed reference value Nf is used m With feedback value Nf e The error between N As input and with the input parameter measured value u m The fan nonlinear speed controller as feedforward calculates the input parameter measurement value deviation and uses the calculated input parameter measurement value deviation Δu e The measured value of the input parameter u m Correction is performed to obtain the corrected engine input parameter measurement value u a =u m +Δu e , and then input it into the aircraft engine onboard adaptive model as the real value of the engine input parameter. The fan nonlinear speed controller is preferably designed using a nonlinear control algorithm to adapt to the strong nonlinear working state of the engine.

[0071] (2) Airborne adaptive model:

[0072] The onboard adaptive model is composed of the second onboard model as the main onboard model and the Kalman filter. Its function is to estimate the performance parameters of the aircraft engine including the health parameters online, so as to realize the full life cycle tracking of the engine, and thus accurately estimate the real thrust, surge margin and other unmeasurable / difficult to measure performance parameters of the engine. Figure 1 As shown, the second airborne model is based on the u obtained in content (1) a As input parameter values, the Kalman filter uses the measured output parameters y of the real engine m The measured output parameters estimated by the second onboard model The normalized residual Δy between norEstimate the degradation of the health parameters of the real engine and obtain the degradation of the health parameters of each engine component Then use The second onboard model is updated to ensure that the second onboard model matches the real engine in real time throughout its life cycle.

[0073] The second airborne model in the above-mentioned airborne adaptive model can adopt the engine nonlinear system dynamic model with health parameter state augmentation. Taking the next generation multivariable large bypass ratio commercial geared turbofan engine as an example, its discrete nonlinear system dynamic model expression is shown as follows:

[0074] x k+1 =f(x k ,u k ,h k )+w k

[0075] y k =g(x k ,u k ,h k )+v k

[0076] z k = l(x k ,u k ,h k )

[0077] Where x is the state variable, u is the control variable, y is the measurable output variable, and z is the difficult / unmeasurable performance parameter. w and v are zero-mean, uncorrelated Gaussian white noises, with w~(0,Q n )(Q n is the system noise covariance matrix), v~(0,R n )(R n is the measurement noise covariance matrix), and h is the component health parameter degradation amount to be estimated.

[0078] In order to obtain the degradation of the engine health parameters through the nonlinear filtering estimation method, the health parameters are expanded into the state quantity of the system, and the dynamic model of the discrete nonlinear system with the expanded state is expressed as follows:

[0079]

[0080] The Kalman filter in the above-mentioned airborne adaptive model can adopt various existing improved Kalman filter algorithms, such as extended Kalman filter, unscented Kalman filter, cubature Kalman filter and particle filter, etc.; or, adopt the latest developed improved spherical unscented Kalman filter iSUKF (Chen Qian et al., A kind of airborne adaptive model of aircraft engine, CN202211397767.3), which effectively takes into account both accuracy and real-time performance, but the real-time performance still needs to be improved. To this end, this embodiment further improves the iSUKF algorithm and proposes an improved spherical unscented Kalman filter estimator based on the minimum number of sigma points (hereinafter referred to as miSUKF). First, by using the principle that the mean and covariance matrix should be equal to the mean and variance matrix of the state quantity, the uncertainty propagation mechanism of the unscented transformation is improved; then, the input-output nonlinear prediction model obtained by offline training of experimental data is used to replace the traditional engine nonlinear component-level model to calculate the output estimation value corresponding to different sigma points; finally, the scaling factor of the propagation of the posterior state estimation error covariance matrix is ​​introduced to ensure the stability of the numerical estimation during the update process. The basic principles and steps of miSUKF are as follows:

[0081] Step 1, k = 0, initialize the posterior state estimate and the posterior state estimation error covariance matrix

[0082]

[0083] Step 2: Perform singular value decomposition SVD:

[0084]

[0085] Among them, U k , S k , They represent the k-time pairs Perform SVD to obtain the unitary matrix, singular value matrix, and transposed matrix of the unitary matrix;

[0086] Step 3: Calculate the weighted coefficient W of the mean and covariance matrix:

[0087]

[0088] Among them, W0, W1, …, W n , are the weight coefficients corresponding to the 1st, 2nd, …, n+1th sigma points respectively, and the parameter vector δ is a non-zero adjustment parameter;

[0089] Step 4: Estimate the posterior state at time k Calculate n+1 sigma points:

[0090]

[0091] in, I is the identity matrix;

[0092] Step 5: Time update:

[0093] According to the known discrete state equation of the aircraft engine nonlinear system f(·,·), n+1 sigma points σ aug,i,k Convert to

[0094]

[0095] σ aug,i,k represents the i-th sigma point at time k, i is the index of the sigma point, i = 0, 1, ..., n, n is the dimension of the state quantity x; u k+1 represents the control input at time k+1; Indicates that according to the i-th sigma point σ aug,i,k and u k+1 The calculated posterior state sample value at time k+1;

[0096] Then the prior state estimate at time k+1 is for:

[0097]

[0098] W i Represents the weight coefficients of the state estimation mean and error covariance matrix;

[0099] Next, we get the prior state estimation error covariance matrix at time k+1: for:

[0100]

[0101] Q k represents the process noise matrix of the system at time k;

[0102] Step 6: Measurement update:

[0103] Step 1: Calculate the estimated output based on the known nonlinear system output function g(·,·)

[0104]

[0105] In order to further improve the real-time performance, g(·,·) here no longer uses the nonlinear component-level model of the engine, but uses the nonlinear prediction model obtained by offline training of test data, such as the Wiener model and the neural network model. By establishing a nonlinear mapping relationship between input and output, the iterative calculation process of the component-level model can be effectively avoided, reducing the amount of calculation.

[0106] Prior output at time k+1 estimate:

[0107]

[0108] Step 2: Considering the measurement noise, the a priori output estimation error covariance matrix at time k+1

[0109]

[0110] The estimated error cross-covariance matrix between the prior state and the output at time k+1

[0111]

[0112] Step 7: Posterior state estimation and the posterior state estimation error covariance matrix renew:

[0113]

[0114]

[0115] in, is the covariance matrix scaling factor, λ i The preferred value range of (i=1,…,n,) is 0.5~1.2, which can be set based on experience;

[0116] Step 8, k=k+1, propagate and Return to step 2 and extract The health parameter degradation amount in is used to update the second onboard model to track the actual engine operating state, thereby ensuring that the second onboard model can accurately calculate the engine performance parameters.

[0117] (3) Health parameter update module:

[0118] In order to ensure that the first airborne model can accurately estimate the measurement deviation of the input parameter value throughout the engine life cycle, the present invention designs a health parameter update module, which uses the health parameters estimated by the airborne adaptive model to update the first airborne model online. The noise and outliers in the first airborne model cause insufficient updates of the health parameters. The filtering algorithm is used to Filtering is performed to obtain noise-free health parameter degradation Secondly, due to the measurement uncertainty of the input parameter value, the first airborne model will dynamically adjust the fan speed in real time through the fan nonlinear speed controller to match the actual engine fan speed. In this process, updating its health parameters will inevitably cause the first airborne model to have an underdetermined control problem, resulting in system instability. For this reason, the present invention does not update the health parameters of the first airborne model in the real engine transient state, but in the steady state, when the engine does not experience performance degradation, the first airborne model fan speed matches the actual engine speed, and the output parameter y is measured. m The measured output parameter y estimated by the first onboard model e The residual r between them will be less than or equal to a very small number ε. Therefore, when r≤ε, it is considered that the real engine has not degraded, and the health parameters of the first airborne model are not updated. At this time, the corrected input parameter measurement value u a Directly give the onboard adaptive model; if r>ε, it is considered that the real engine is degraded and must be used The health parameters of the first airborne model are updated, and the new input parameter measurement value deviation is recalculated through the fan nonlinear speed controller until r≤ε is satisfied, and the real input parameter measurement value deviation is obtained, and u is corrected. m ; Finally, u a Give the onboard adaptive model as input parameter values.

[0119] The above filtering algorithm can adopt the existing sliding average method, median method, Savitzky-Golay algorithm, etc. Filtering should not only eliminate noise and outliers in the process of health parameter estimation, but also avoid the problem of data delay caused by filtering and resulting in untimely updates; therefore, the present invention designs an outlier-median average filtering algorithm to Perform filtering; for the current k moment, the filtering process is as follows:

[0120] S1. Obtain the health parameter data set D(t,h) estimated by the onboard adaptive model at the N moments before the current moment;

[0121] S2, for each data point t in the data set D(t,h) i , set an internal window and set t i Compared with all other data points in this window j Perform triple cubic weight function calculation to obtain weight ζ ij :

[0122]

[0123] Wherein, d is the width of the internal window;

[0124] S3. In the internal window of each data point, a polynomial is fitted using the weighted least squares method. The goal is to minimize the weighted residual sum of squares to obtain a new data set Data wnd :

[0125]

[0126] Among them, p(t i ; β) is a polynomial model, β is the coefficient of the polynomial, h j is the jth data of health parameter;

[0127] S4. Discard Data wnd The minimum and maximum values ​​in the filter are averaged to obtain the degradation amount of the health parameter at the current moment after filtering.

[0128]

[0129] The self-correction process of the aircraft engine onboard adaptive model input parameters in this embodiment is as follows: Figure 2 As shown, the details are as follows:

[0130] Step 1: Assume that at the current time k, the true value of the engine input parameter is u r , the fuel flow rate measured by the engine fuel sensor is u m , the engine measurement output parameter is y m ;

[0131] Step 2: Track the actual engine fan speed according to the fan nonlinear speed controller of the first airborne model and calculate Δu e ;

[0132] Step 3: Using Δu e Calculate u a :u a =u m +Δu e ;

[0133] Step 4: Determine whether the real engine is in steady state: if so, go to Step 5; if not, turn off the switch and go to Step 6;

[0134] Step 5: Determine the engine measurement output parameter y m The output parameter y of the first airborne model eIs the error r less than or equal to ε (set to 10 in this embodiment)? -3 ):

[0135] If so, then u e is the actual engine fuel flow, the switch is disconnected, there is no need to correct the health parameters of the first airborne model, and u e to the second airborne model in the airborne adaptive model; at this time, the thrust, surge margin and other difficult / impossible performance parameter values ​​estimated by the airborne adaptive model are the real values ​​of the engine;

[0136] If not, it indicates that the engine has experienced component performance degradation and the first onboard model no longer matches the real engine:

[0137] Step 5-1: At this time, the switch is disconnected, and the onboard adaptive model outputs y according to the actual engine measurement. m With the second onboard model the estimated output The normalized measurement residual Δy nor Real-time estimation of the degradation of the health parameters of the real engine

[0138] Step 5-2: Processing using the outlier-median average filtering algorithm get

[0139] Step 5-3: The switch is closed and the Correct the first airborne model. The fan nonlinear speed controller in the first airborne model module will adjust the fan speed in real time until y m With y e The error r≤ε, at this time, the deviation calculated by the nonlinear speed controller of the first airborne model is the actual measurement deviation of the input parameter;

[0140] Step 6: The corrected input parameter measurement value u a Give the airborne adaptive model;

[0141] Step 7: The onboard adaptive model estimates the performance parameters of the engine that are difficult to measure / impossible to measure; Step 8: Is it finished? If not, jump to Step 1 and continue to the next moment; if yes, the engine operation is finished.

Claims

1. A method for self-correcting input parameters of an aircraft engine onboard adaptive model, wherein the aircraft engine onboard adaptive model comprises a main onboard model and a Kalman filter, and is used for online estimation of aircraft engine performance parameters including health parameters; characterized in that: The input parameter measurement value of the aircraft engine onboard adaptive model is corrected as follows using an additional auxiliary airborne model and a fan nonlinear speed controller: the fan speed measured in real time by the real engine is used as a reference value, the fan speed estimated by the auxiliary airborne model is used as a feedback value, the input parameter measurement value deviation is calculated by a fan nonlinear speed controller that takes the error between the fan speed reference value and the feedback value as input and takes the input parameter measurement value as a feedforward quantity, and the input parameter measurement value is corrected by the calculated input parameter measurement value deviation and then input into the aircraft engine onboard adaptive model; and when the real aircraft engine is in a steady state and the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is greater than a preset error threshold, the health parameter degradation amount of the real engine estimated by the aircraft engine onboard adaptive model is filtered, and then the auxiliary airborne model is updated with the filtered health parameter degradation amount until the error between the measured output parameter of the real aircraft engine and the output parameter of the auxiliary airborne model is less than or equal to the error threshold.

2. The method for self-correcting input parameters of an aircraft engine onboard adaptive model according to claim 1, characterized in that: The filtering process is performed using the following outlier-median average filtering algorithm: S1. Obtain the health parameter data set D(t,h) estimated by the onboard adaptive model at the N moments before the current moment; S2, for each data point t in the data set D(t,h) i , set an internal window and set t i Compared with all other data points in this window j Perform triple cubic weight function calculation to obtain weight ζ ij : Wherein, d is the width of the internal window; S3. In the internal window of each data point, a polynomial is fitted using the weighted least squares method. The goal is to minimize the weighted residual sum of squares to obtain a new data set Data wnd : Among them, p(t i ; β) is a polynomial model, β is the coefficient of the polynomial, h j is the jth data of health parameter; S4. Discard Data wnd The minimum and maximum values ​​in the filter are averaged to obtain the degradation amount of the health parameter at the current moment after filtering.

3. The method for self-correcting input parameters of an aircraft engine onboard adaptive model according to claim 1, characterized in that: The Kalman filter is an improved spherical unscented Kalman filter estimator based on the minimum sigma point number, and the working process is as follows: Step 1, k = 0, initialize the posterior state estimate and the posterior state estimation error covariance matrix Step 2: Perform singular value decomposition SVD: Among them, U k , S k , They represent the k-time pairs Perform SVD to obtain the unitary matrix, singular value matrix, and transposed matrix of the unitary matrix; Step 3: Calculate the weighted coefficient W of the mean and covariance matrix: Among them, W0, W1, …, W n , are the weighting coefficients corresponding to the 1st, 2nd, …, n+1th sigma points respectively, and the parameter vector δ is a non-zero adjustment parameter; Step 4: Estimate the posterior state at time k Calculate n+1 sigma points: in, I is the identity matrix; Step 5: Time update: According to the known discrete state equation f(·,·) of the aircraft engine nonlinear system, n+1 sigma points σ aug,i,k Convert to σ aug,i,k represents the i-th sigma point at time k, i is the index of the sigma point, i = 0, 1, ..., n, n is the dimension of the state quantity x; u k+1 represents the control input at time k+1; Indicates that according to the i-th sigma point σ aug,i,k and u k+1 The calculated posterior state sample value at time k+1; Then the prior state estimate at time k+1 is for: W i Represents the weight coefficients of the state estimation mean and error covariance matrix; Next, we get the prior state estimation error covariance matrix at time k+1: for: Q k represents the process noise matrix of the system at time k; Step 6: Measurement update: Step 1: Calculate the estimated output based on the known nonlinear system output function g(·,·) Prior output at time k+1 estimate: Step 2: Considering the measurement noise, the a priori output estimation error covariance matrix at time k+1 The estimated error cross-covariance matrix between the prior state and the output at time k+1 Step 7: Posterior state estimation and the posterior state estimation error covariance matrix renew: in, is the covariance matrix scaling factor; Step 8, k=k+1, propagate and Return to step 2 and extract The degradation amount of health parameters in is used to update the host onboard model.

4. The method for self-correcting input parameters of an aircraft engine onboard adaptive model according to claim 3, characterized in that: The nonlinear system output function g(·,·) is a nonlinear prediction model obtained by offline training using test data.

5. An aircraft engine onboard adaptive model input parameter self-correction device, the aircraft engine onboard adaptive model includes a main onboard model and a Kalman filter, and is used for online estimation of aircraft engine performance parameters including health parameters; characterized in that: The device includes an additional auxiliary airborne model, a fan nonlinear speed controller and a health parameter update module; The auxiliary airborne model and the fan nonlinear speed controller are used to correct the input parameter measurement value of the aircraft engine airborne adaptive model as follows: using the fan speed measured in real time by the real engine as a reference value, using the fan speed estimated by the auxiliary airborne model as a feedback value, calculating the input parameter measurement value deviation through a fan nonlinear speed controller that takes the error between the fan speed reference value and the feedback value as input and the input parameter measurement value as a feedforward quantity, and correcting the input parameter measurement value with the calculated input parameter measurement value deviation and then inputting it into the aircraft engine airborne adaptive model; The health parameter update module is used to filter the health parameter degradation amount of the real engine estimated by the aircraft engine onboard adaptive model when the real aircraft engine is in a steady state and the error between the measured output parameters of the real aircraft engine and the output parameters of the auxiliary airborne model is greater than a preset error threshold, and then use the filtered health parameter degradation amount to update the auxiliary airborne model until the error between the measured output parameters of the real aircraft engine and the output parameters of the auxiliary airborne model is less than or equal to the error threshold.

6. The aircraft engine onboard adaptive model input parameter self-correction device as claimed in claim 5, characterized in that: The filtering process is performed using the following outlier-median average filtering algorithm: S1. Obtain the health parameter data set D(t,h) estimated by the onboard adaptive model at the N moments before the current moment; S2, for each data point t in the data set D(t,h) i , set an internal window and set t i Compared with all other data points in this window j Perform triple cubic weight function calculation to obtain weight ζ ij : Wherein, d is the width of the internal window; S3. In the internal window of each data point, a polynomial is fitted using the weighted least squares method. The goal is to minimize the weighted residual sum of squares to obtain a new data set Data wnd : Among them, p(t i ; β) is a polynomial model, β is the coefficient of the polynomial, h j is the jth data of health parameter; S4. Discard Data wnd The minimum and maximum values ​​in the filter are averaged to obtain the degradation amount of the health parameter at the current moment after filtering.

7. The aircraft engine onboard adaptive model input parameter self-correction device as claimed in claim 5, characterized in that: The Kalman filter is an improved spherical unscented Kalman filter estimator based on the minimum sigma point number, and the working process is as follows: Step 1, k = 0, initialize the posterior state estimate and the posterior state estimation error covariance matrix Step 2: Perform singular value decomposition SVD: Among them, U k , S k , They represent the k-time pairs Perform SVD to obtain the unitary matrix, singular value matrix, and transposed matrix of the unitary matrix; Step 3: Calculate the weighted coefficient W of the mean and covariance matrix: Among them, W0, W1, …, W n , are the weighting coefficients corresponding to the 1st, 2nd, …, n+1th sigma points respectively, and the parameter vector δ is a non-zero adjustment parameter; Step 4: Estimate the posterior state at time k Calculate n+1 sigma points: in, I is the identity matrix; Step 5: Time update: According to the known discrete state equation f(·,·) of the aircraft engine nonlinear system, n+1 sigma points σ aug,i,k Convert to σ aug,i,k represents the i-th sigma point at time k, i is the index of the sigma point, i = 0, 1, ..., n, n is the dimension of the state quantity x; u k+1 represents the control input at time k+1; Indicates that according to the i-th sigma point σ aug,i,k and u k+1 The calculated posterior state sample value at time k+1; Then the prior state estimate at time k+1 is for: W i Represents the weight coefficients of the state estimation mean and error covariance matrix; Next, we get the prior state estimation error covariance matrix at time k+1: for: Q k represents the process noise matrix of the system at time k; Step 6: Measurement update: Step 1: Calculate the estimated output based on the known nonlinear system output function g(·,·) Prior output at time k+1 estimate: Step 2: Considering the measurement noise, the a priori output estimation error covariance matrix at time k+1 The estimated error cross-covariance matrix between the prior state and the output at time k+1 Step 7: Posterior state estimation and the posterior state estimation error covariance matrix renew: in, is the covariance matrix scaling factor; Step 8, k=k+1, propagate and Return to step 2 and extract The degradation amount of health parameters in is used to update the host onboard model.

8. The aircraft engine onboard adaptive model input parameter self-correction device as claimed in claim 7, characterized in that: The nonlinear system output function g(·,·) is a nonlinear prediction model obtained by offline training using test data.

9. An aircraft engine control system, comprising an aircraft engine onboard adaptive model, wherein the aircraft engine onboard adaptive model comprises a main onboard model and a Kalman filter, and is used for online estimation of aircraft engine performance parameters including health parameters; characterized in that: The aircraft engine control system further comprises an aircraft engine onboard adaptive model input parameter self-correction device as claimed in any one of claims 5 to 8.

Images