Aero-engine on-wing thrust evaluation method

Abstract

The invention discloses an on-wing thrust assessment method for an aero-engine. The method comprises the following steps: step 1, acquiring ground reference data, historical flight monitoring data and flight monitoring data to be assessed, and respectively converting the data into a standard environment; step 2, constructing a neural network model, taking state parameters of the converted ground reference data as input, and taking a turbine rear key parameter conversion value and a thrust parameter conversion value as output; 3, screening an optimal test model through one-time feature migration, that is, after a plurality of instance models are trained, selecting a model with the minimum deviation in combination with historical flight monitoring data; 4, thrust evaluation is conducted through secondary feature migration, flight monitoring data to be evaluated are input into the optimal test model, a correction coefficient is calculated based on linear regression, a thrust prediction value is calibrated, and finally the engine physical thrust is obtained. The method can effectively solve the difference between ground and high-altitude working conditions, and achieves the precise quantitative evaluation of the on-wing thrust of the aero-engine.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aero-engine maintenance support, and particularly relates to an in-wing thrust evaluation method of an aero-engine. BACKGROUND

[0002] The aero-engine thrust is a core performance parameter for evaluating the performance of a fighter and ensuring the execution efficiency of a task, and accurate acquisition of the aero-engine thrust plays a key supporting role in engine combat selection, performance degradation dynamic monitoring and maintenance support decision-making. The in-wing thrust specifically refers to the actual thrust generated by the aero-engine installed on the wing (or fuselage) of the fighter in a flight state, which is different from the test thrust of the aero-engine on the ground test bench, and can directly reflect the real working performance of the aero-engine in a high-altitude complex environment (such as different altitudes, temperatures and Mach numbers) and an integral assembly state, and is a core index for measuring the air maneuverability and load carrying capacity of the fighter.

[0003] At present, direct measurement of the aero-engine thrust can only be achieved in a ground environment: with the aid of a ground test bench and a high-precision sensor and a force measuring device, the test thrust of the aero-engine separated from the integral machine can be accurately tested, and the technology is relatively mature. However, in the actual application in the field, the acquisition of the in-wing thrust faces significant bottlenecks: on the one hand, the high-altitude environment has problems such as severe temperature fluctuations, large pressure gradient changes and complex airflow interference, and it is difficult to deploy test equipment of the ground test bench level; on the other hand, during the flight of the fighter, the aero-engine is deeply coupled with the aerodynamic layout of the integral machine, the fuel supply system and the avionics control system, and it is impossible to obtain the thrust data by simply disassembling and testing. Therefore, the current field judgment of the in-wing thrust still mainly relies on the subjective perception of the pilot (such as acceleration response and climb rate), lacks quantitative data support, and seriously restricts the accurate evaluation of the air performance state of the aero-engine.

[0004] The existing technical solutions for indirectly acquiring the thrust cannot effectively solve the problem of in-wing thrust evaluation, and the specific limitations are as follows:

[0005] Firstly, the traditional physical modeling method needs to accurately model the aero-engine, but the field engine measurement points are limited, real-time structural data is missing, and it is difficult to establish a complete model. Chinese patent CN110686813A discloses a method for identifying the engine thrust of a twin-engine aircraft through a ground sliding test, and the thrust is solved by using a simplified physical motion equation. The disadvantage is that it can only be applied to the sliding stage, and the air thrust still cannot be calculated.

[0006] Second, based on the data-driven method, Chinese patent CN112149233A discloses an aero-engine dynamic thrust estimation method based on an echo state network, a simple artificial intelligence model is established to train a thrust estimator, the disadvantage is that the influence of different external environments (flight altitude, temperature, Mach number, etc.) on input parameters is not considered, and it is only suitable for thrust calculation under the same environment. Chinese patent CN111860791A discloses an aero-engine thrust estimation method and device based on similarity transformation, which utilizes the similarity principle to convert engine parameters, and utilizes a data-driven model to realize thrust calculation on new parameters, the disadvantage is that the non-linear deviation between different working conditions is not considered, and the high-altitude environment and the ground environment cannot be accurately equivalent based on simple similarity conversion.

[0007] In summary, the prior art cannot break through the working condition barrier between the ground and the high altitude, and cannot realize the precise quantitative calculation of the in-flight thrust of the aero-engine, therefore, it is particularly necessary to provide a technical solution capable of adapting to the flight environment in the field and effectively obtaining the in-flight thrust. SUMMARY

[0008] The purpose of the present application is to provide an aero-engine in-flight thrust evaluation method which can effectively obtain the in-flight thrust.

[0009] The technical solution adopted by the present application is as follows:

[0010] An aero-engine in-flight thrust evaluation method, the method comprising the following steps:

[0011] Step 1: data acquisition and equivalent conversion: acquiring ground reference data, historical flight monitoring data and to-be-evaluated flight monitoring data of the aero-engine, wherein the ground reference data includes state parameters representing the running state of the engine and thrust parameters as evaluation targets, the historical flight monitoring data and the to-be-evaluated flight monitoring data include state parameters representing the running state of the engine; the ground reference data, the historical flight monitoring data and the to-be-evaluated flight monitoring data are respectively converted to standard environmental conditions according to a preset rule, to obtain converted ground reference data, historical flight monitoring data and to-be-evaluated flight monitoring data;

[0012] Step 2: data-driven model construction: constructing a neural network model according to the engine structure, taking the state parameters in the converted ground reference data as the model input, and taking the converted values of the key parameters after the turbine and the converted values of the thrust parameters as the model output;

[0013] Step 3: One-time feature migration: based on the converted ground reference data, the training set and the validation set are divided, and the neural network model is trained for multiple rounds to obtain multiple instance models. The performance of the instance models on the validation set is screened by a preset loss index, and the instance model with the best performance is selected as the optimal test model.

[0014] Step 4: Secondary feature migration: input the converted flight monitoring data into the optimal test model to obtain the optimal prediction value of the thrust parameter conversion value and the optimal prediction value of the turbine key parameter conversion value output by the optimal test model. Based on the linear regression algorithm, the optimal prediction value of the turbine key parameter conversion value output by the optimal test model and the measured turbine key parameter conversion value in the converted flight monitoring data are used to calculate a correction coefficient. The optimal prediction value of the thrust parameter conversion value is calibrated by the correction coefficient to obtain the final in-flight engine thrust conversion value final prediction value. Based on the thrust conversion formula, the final in-flight engine thrust conversion value final prediction value is converted into the final engine physical thrust prediction value.

[0015] Further, the state parameters in step 1 include control parameters, air path cross-section parameters, and variable geometry parameters. The control parameters include high-pressure rotor speed N1, low-pressure rotor speed N2, and throttle lever angle PLA. The air path cross-section parameters include fan inlet total temperature T1, fan inlet total pressure P1, compressor outlet total pressure P 31 , low-pressure turbine outlet total pressure P6, and low-pressure turbine outlet total temperature T6. The variable geometry parameters include fan guide vane angle a1, compressor guide vane angle a2, and engine tail nozzle angle D8. The thrust parameter is the engine physical thrust F. The turbine key parameter is the low-pressure turbine outlet total pressure P6 or the low-pressure turbine outlet total temperature T6 in the air path cross-section parameters.

[0016] Further, the ground reference data and the flight monitoring data are converted to standard environmental conditions according to a preset rule, and the specific conversion process of the high-pressure rotor speed N1, the low-pressure rotor speed N2, the compressor outlet total pressure P 31 , the low-pressure turbine outlet total pressure P6, the low-pressure turbine outlet total temperature T6, and the engine physical thrust F is as follows: The high-pressure rotor speed conversion formula is as follows: ; Where: N1R X represents the converted high-voltage rotor speed, N1 represents the measured high-voltage rotor speed, and X represents the calculated high-voltage rotor speed. N1 T1 is the high-voltage rotor speed correction factor, 288.15 is the standard ambient reference temperature, T1 is the measured total temperature at the fan inlet, and 273.15 is the temperature conversion factor. The formula for converting low-pressure rotor speed is as follows: ; Where: N 2R X is the converted low-pressure rotor speed, N2 is the measured low-pressure rotor speed, and X is the calculated low-pressure rotor speed. N2 T1 is the low-pressure rotor speed correction factor, 288.15 is the standard ambient reference temperature, T1 is the measured total temperature at the fan inlet, and 273.15 is the temperature conversion factor. The formula for converting the total outlet pressure of the compressor is as follows: ; Where: P 31R P is the converted total compressor outlet pressure. 31 P1 represents the measured total pressure at the compressor outlet, 101.325 represents the standard ambient reference pressure, and P1 represents the measured total pressure at the fan inlet. The formula for converting total pressure after low-pressure turbine is as follows: ; Where: P 6R P6 is the converted total pressure after the low-pressure turbine, 101.325 is the standard ambient reference pressure, and P1 is the measured total pressure at the fan inlet. The formula for calculating the total temperature after the low-pressure turbine is as follows: ; Wherein: T 6R T6 represents the converted total temperature after the low-pressure turbine, and X represents the measured total temperature after the low-pressure turbine. T6 273.15 is the total temperature correction factor after the low-pressure turbine, and 273.15 is the temperature conversion factor. The formula for converting thrust parameters is as follows: ; Wherein: F R Here are the converted thrust parameters, where F is the measured physical thrust of the engine, and X is... F P1 is the thrust correction factor, 101.325 is the standard environmental reference pressure, and P1 is the measured total pressure at the fan inlet.

[0017] Furthermore, the neural network model mentioned in step 2 is a long short-term memory neural network model, which includes a parameter input module, a feature extraction module, a feature coupling module, a feature dimensionality reduction module, and a result output module; the specific working process of each module is as follows: The parameter input module splices l pieces of continuous time sequence input parameter data into a single input sample, where l is the number of continuous time sequence data; The feature extraction module extracts features of the input sample and outputs a feature vector with a size of [l, c1], where c1 is a preset hidden feature dimension; The feature coupling module adopts a bidirectional long short-term memory neural network model to perform bidirectional information coupling on the feature vector and outputs a coupled feature vector with a length of [l, c2], where c2 is a preset hidden feature dimension of the bidirectional long short-term memory neural network model; The feature dimension reduction module adopts a fully connected neural network to perform flattening and dimension reduction processing on the coupled feature vector; The result output module outputs a two-dimensional result . The result output module outputs a two-dimensional result , which includes a predicted value of a total pressure conversion value after a low-pressure turbine output by the model and a predicted value of a thrust parameter conversion value output by the model .

[0018] Further, in step 3, the preset loss index is a loss function value Loss Val calculated on the validation set Val , and the calculation formula of the loss function value Loss 6R,Val is as follows: ; , where mse() is a mean square error calculation function, P R,Val is a measured total pressure conversion value after a low-pressure turbine in the validation set, is a predicted value of a total pressure conversion value after a low-pressure turbine output by the model, F Val is a measured thrust parameter conversion value in the validation set, is a predicted value of a thrust parameter conversion value output by the model.

[0019] Further, in step 3, based on the converted ground reference data, a training set and a validation set are divided, and the neural network model is trained for multiple rounds to obtain multiple instance models, and the specific process of selecting part of the instance models with the best performance on the validation set based on the preset loss index is as follows: The converted ground reference data is divided into a training set and a validation set, and the neural network model is trained for w rounds, each round generating an instance model, and w instance models are obtained; the loss function value Loss Val of the w instance models on the validation set is calculated, and m instance models with the smallest loss function value Loss Val are selected, where 2≤m

[0020] Further, in step 3, the converted historical flight monitoring data is input into the screened instance model as auxiliary domain data, and the deviation between the converted key parameter conversion value after the turbine output by each instance model based on the converted historical flight monitoring data and the measured key parameter conversion value after the turbine in the historical flight monitoring data is compared, and the instance model with the smallest deviation is selected as the specific process of the optimal test model as follows: The converted historical flight monitoring data is input into the above-mentioned m instance models, the low-pressure turbine total pressure conversion value prediction value based on the converted historical flight monitoring data output by each instance model is obtained , and the instance model with the smallest mean square error value between the low-pressure turbine total pressure conversion value P 6R,Hist measured in the historical flight monitoring data is calculated as the optimal test model.

[0021] Further, the specific process of step 4 is as follows: Step 4.1: Constructing a linear relationship formula based on a linear regression algorithm: input the converted to-be-evaluated flight monitoring data into the optimal test model, and constructing a linear relationship formula based on the optimal prediction value of the key parameter conversion value after the turbine output by the optimal test model and the measured key parameter conversion value after the turbine in the converted to-be-evaluated flight monitoring data, the linear relationship formula is as follows: ; Wherein, P 6R,Real is the measured low-pressure turbine total pressure conversion value in the to-be-evaluated flight monitoring data, is the optimal prediction value of the key parameter conversion value after the turbine output by the optimal test model; h1, h2 are correction coefficients to be solved; δ is the error of linear transformation; Step 4.2: defining the matrix and vector required for solving the correction coefficient: Suppose the number of to-be-evaluated flight monitoring data samples required for calculating the correction coefficient is n, then: Input matrix Z: the measured low-pressure turbine total pressure conversion value P 6R,Real of n to-be-evaluated flight monitoring data samples and the constant 1 are respectively taken as two columns to form a matrix, and the matrix expression is: ; Coefficient vector H: composed of correction coefficients h1, h2 to be solved, and the vector expression is ; Output vector Y: composed of n to-be-evaluated flight monitoring data samples corresponding to , and the vector expression is: ; Error vector δ: composed of the linear transformation errors of n to-be-evaluated flight monitoring data samples, and the vector expression is ; Step 4.3: Derive the solution formula of correction coefficient and calculate the coefficient vector H: Convert the linear relationship formula in step 4.1 into matrix form Y = ZH + δ; To minimize the sum of squares of error vector δ, the solution formula of correction coefficient vector H is derived by least square method: ; Wherein, Z T is the transpose matrix of input matrix Z, which is obtained by interchanging the rows and columns of input matrix Z; Z T Z is the transpose matrix Z T is the product matrix of input matrix Z, which is calculated according to the matrix multiplication rule; (Z T Z) −1 is the inverse matrix of matrix ZTZ, which is calculated by matrix inversion operation; Substitute the input matrix Z and output vector Y defined in step 4.2 into the above solution formula to calculate the coefficient vector H; Step 4.4: Extract correction coefficients and calibrate the optimal prediction value of thrust parameter conversion value: From the coefficient vector H calculated in step 4.3, extract the first element as correction coefficient h1 and the second element as correction coefficient h2; Input the converted aero-engine flight monitoring data to be evaluated into the optimal test model to obtain the optimal prediction value of thrust parameter conversion value output by the optimal test model ; Calibrate the optimal prediction value of thrust parameter conversion value according to the calibration formula to obtain the final prediction value of aero-engine in-flight thrust conversion value ; The calibration formula is: ; Wherein is the final prediction value of aero-engine in-flight thrust conversion value, is the optimal prediction value of thrust parameter conversion value output by the optimal test model, and h1 and h2 are correction coefficients; The calculation formula of the final prediction value of engine physical thrust according to the thrust parameter conversion formula is as follows: ; Wherein: is the final prediction value of engine physical thrust, is the final prediction value of aero-engine in-flight thrust conversion value, and X F is the thrust correction coefficient, 101.325 is the standard environmental reference pressure, and P1 is the fan inlet total pressure.

[0022] In summary, by adopting the technical scheme, the application has the beneficial effects of:

[0023] The application first converts the ground reference data, historical flight monitoring data and to-be-evaluated flight monitoring data to a standard environmental condition, eliminating the influence of environmental differences on parameters; then filters an optimal test model of the historical flight monitoring data through a first feature migration, solving the problem of incompatibility between the ground-trained neural network model and high-altitude working conditions; finally, a correction coefficient is calculated based on a linear regression algorithm through a second feature migration, the optimal prediction value of the thrust parameter conversion value is calibrated to compensate for the deviation between the model output and the measured value, and the final prediction value of the in-flight thrust conversion value of the aero-engine is converted into the final prediction value of the engine physical thrust. The application breaks through the working condition barriers between the ground test environment and the high-altitude flight environment, and can realize the quantitative evaluation of the in-flight thrust of the aero-engine through data processing and model optimization without relying on large-scale test equipment that cannot be deployed in the field, solving the defects of the prior art that can only measure ground thrust and rely on subjective perception, and providing precise data support for engine in-flight performance evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 The flowchart of the application. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical scheme and advantages of the embodiments of the application clearer, the technical scheme in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. The components of the embodiments of the application described and shown in the drawings can be arranged and designed in various different configurations.

[0026] Therefore, the following detailed description of the embodiments of the application provided in the drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the application without creative labor are within the scope of protection of the application.

[0027] It should be noted that the embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0028] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0029] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship commonly used when the product of the present application is used, or the orientation or positional relationship commonly understood by those skilled in the art, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second" and the like are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0030] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "arrangement", "installation", "connection", "connection" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be connected inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0031] As shown in Figure 1 The present application discloses an aero-engine in-wing thrust evaluation method, the method comprising the following steps:

[0032] Step 1: data acquisition and equivalent conversion: obtaining ground reference data, historical flight monitoring data and to-be-evaluated flight monitoring data of an aero-engine, wherein the ground reference data includes state parameters representing the operating state of the engine and thrust parameters as evaluation targets, and the historical flight monitoring data and the to-be-evaluated flight monitoring data include state parameters representing the operating state of the engine; the ground reference data, the historical flight monitoring data and the to-be-evaluated flight monitoring data are converted to standard environmental conditions according to a preset rule, to obtain converted ground reference data, historical flight monitoring data and to-be-evaluated flight monitoring data;

[0033] Step 2: data-driven model construction: constructing a neural network model according to the structure of the engine, taking the state parameters in the converted ground reference data as model inputs, and taking the converted values of the key parameters after the turbine and the converted values of the thrust parameters as model outputs;

[0034] Step 3: One-time feature migration: based on the converted ground reference data, the training set and the validation set are divided, and the neural network model is trained for multiple rounds to obtain multiple instance models, and the performance of the instance models on the validation set is screened out by a preset loss index; the converted historical flight monitoring data is input into the screened instance model as auxiliary domain data, and the deviation between the converted historical flight monitoring data and the measured turbine key parameter conversion value in the historical flight monitoring data is compared, and the instance model with the smallest deviation is selected as the optimal test model;

[0035] Step 4: Secondary feature migration: input the converted to-be-evaluated flight monitoring data into the optimal test model to obtain the optimal prediction value of the thrust parameter conversion value and the optimal prediction value of the turbine key parameter conversion value output by the optimal test model; based on a linear regression algorithm, the optimal prediction value of the turbine key parameter conversion value output by the optimal test model and the measured turbine key parameter conversion value in the converted to-be-evaluated flight monitoring data are used to calculate a correction coefficient; the optimal prediction value of the thrust parameter conversion value is calibrated by the correction coefficient to obtain the final in-flight engine thrust conversion value final prediction value; based on a thrust conversion formula, the in-flight engine thrust conversion value final prediction value is converted into the engine physical thrust final prediction value.

[0036] The ground reference data, the historical flight monitoring data and the to-be-evaluated flight monitoring data are first converted to standard environmental conditions to eliminate the influence of environmental differences on the parameters; then the optimal test model suitable for the historical flight monitoring data is selected through one-time feature migration to solve the problem of incompatibility between the ground-trained neural network model and high-altitude working conditions; finally, the correction coefficient is calculated based on a linear regression algorithm through secondary feature migration to calibrate the optimal prediction value of the thrust parameter conversion value to compensate for the deviation between the model output and the measured value, and the in-flight engine thrust conversion value final prediction value is converted into the engine physical thrust final prediction value. The application breaks through the working condition barrier between the ground test environment and the high-altitude flight environment, and can realize the quantitative evaluation of the in-flight engine thrust without relying on large test equipment that cannot be deployed in the field, through data processing and model optimization, solves the defects that the prior art can only measure the ground thrust and relies on subjective perception, and provides precise data support for engine in-flight performance evaluation.

[0037] Further, the state parameters in step 1 include control parameters, air path cross-section parameters and variable geometry parameters; The control parameters include high-pressure rotor speed N1, low-pressure rotor speed N2 and throttle lever angle PLA; The air path cross-section parameters include fan inlet total temperature T1, fan inlet total pressure P1, compressor outlet total pressure P 31 , low-pressure turbine outlet total pressure P6 and low-pressure turbine outlet total temperature T6. The variable geometry parameters include fan vane angle a1, compressor vane angle a2, engine nozzle angle D8; The thrust parameter is engine physical thrust F; The turbine rear key parameter is low pressure turbine rear total pressure P6 or low pressure turbine rear total temperature T6 in the gas path section parameter.

[0038] Further, the ground reference data and the flight monitoring data are converted to standard environmental conditions according to preset rules, wherein the high pressure rotor speed N1, the low pressure rotor speed N2, the compressor outlet total pressure P 31 , the low pressure turbine rear total pressure P6, the low pressure turbine rear total temperature T6, and the engine physical thrust F are converted as follows: The high pressure rotor speed conversion formula is as follows: ; Wherein: N 1R is the converted high pressure rotor speed, N1 is the measured high pressure rotor speed, X N1 is the high pressure rotor speed correction coefficient, 288.15 is the standard environmental reference temperature, T1 is the measured fan inlet total temperature, and 273.15 is the temperature conversion coefficient; The low pressure rotor speed conversion formula is as follows: ; Wherein: N 2R is the converted low pressure rotor speed, N2 is the measured low pressure rotor speed, X N2 is the low pressure rotor speed correction coefficient, 288.15 is the standard environmental reference temperature, T1 is the measured fan inlet total temperature, and 273.15 is the temperature conversion coefficient; The compressor outlet total pressure conversion formula is as follows: ; Wherein: P 31R is the converted compressor outlet total pressure, P 31 is the measured compressor outlet total pressure, 101.325 is the standard environmental reference pressure, and P1 is the measured fan inlet total pressure; The low pressure turbine rear total pressure conversion formula is as follows: ; Wherein: P 6R is the converted low pressure turbine rear total pressure, P6 is the measured low pressure turbine rear total pressure, 101.325 is the standard environmental reference pressure, and P1 is the measured fan inlet total pressure; The low pressure turbine rear total temperature conversion formula is as follows: ; Wherein: T 6RT6 is the converted low-pressure turbine total temperature, T6 is the measured low-pressure turbine total temperature, and X is the low-pressure turbine total temperature correction coefficient T6 T6 is the converted low-pressure turbine total temperature, T6 is the measured low-pressure turbine total temperature, and X is the low-pressure turbine total temperature correction coefficient The thrust parameter conversion formula is as follows: ; Wherein: F R T6 is the converted low-pressure turbine total temperature, T6 is the measured low-pressure turbine total temperature, and X is the low-pressure turbine total temperature correction coefficient F T6 is the converted low-pressure turbine total temperature, T6 is the measured low-pressure turbine total temperature, and X is the low-pressure turbine total temperature correction coefficient

[0039] The application provides specific conversion formulas for core parameters such as high-pressure rotor speed, low-pressure turbine total pressure, and thrust, and by introducing correction coefficients (X N1 , X N2 , X T6 , X) and standard environmental reference values (288.15K, 101.325kPa), the measured parameters under different environments are accurately converted to a unified standard environment, eliminating the interference of environmental factors such as temperature and pressure on the parameters. The "environmental normalization" of ground reference data and flight monitoring data is realized, avoiding the data deviation caused by uncorrected environmental parameters in the prior art, providing unified dimension parameters for subsequent model training (based on ground data) and flight data adaptation (one-time migration), ensuring the applicability of the model in cross-environment scenarios, and improving the accuracy of the basic data for in-flight thrust evaluation.

[0040] Further, the neural network model in step 2 is a long short-term memory neural network model, which includes a parameter input module, a feature extraction module, a feature coupling module, a feature dimension reduction module and a result output module. The working process of each module is as follows: The parameter input module concatenates l pieces of continuous time series input parameter data into a single input sample, where l is the number of continuous time series data; The feature extraction module extracts features from the input sample and outputs a feature vector with a size of [l, c1], where c1 is a preset hidden feature dimension; The feature coupling module uses a bidirectional long short-term memory neural network model to perform bidirectional information coupling on the feature vector, and outputs a coupled feature vector with a length of [l, c2], where c2 is a preset hidden feature dimension of the bidirectional long short-term memory neural network model; The feature dimension reduction module uses a fully connected neural network to perform flattening and dimension reduction processing on the coupled feature vector; The result output module outputs a two-dimensional result ; which includes the low-pressure turbine total pressure conversion value predicted by the model and the thrust parameter conversion value predicted value output by the model .

[0041] In specific implementation, the number of continuous time series data in the parameter input module is 6, that is, 6 continuous time series of input parameter data are spliced to form a single input sample; the preset hidden feature dimension c1 of the feature extraction module is 4, and after feature extraction of the input sample, a feature vector with a size of [6, 4] is output, and the preset hidden feature dimension c1=4 is an artificial set hyperparameter; the preset hidden feature dimension c2 of the bidirectional long short-term memory neural network adopted by the feature coupling module is 32; when processing the feature vector, the module considers both the information of forward transmission and the information of reverse transmission between samples, and the forward transmission and the reverse transmission process will output hidden features once, so the finally output coupled feature vector has a length of [6, 64] (64=2×32); the feature dimension reduction module performs flattening processing on the coupled feature vector with a length of [6, 64] and inputs it into the regression module composed of a fully connected neural network to complete dimension reduction; after dimension reduction processing, the result output module outputs a two-dimensional result .

[0042] The present application solves the problem that the existing data-driven model ignores the time sequence characteristics and has insufficient dynamic fitting capability, and improves the accuracy of the model in describing the dynamic running state of the engine through the time sequence information coupling of the bidirectional LSTM and the multi-module cooperation.

[0043] Further, in step 3: the preset loss index is the loss function value Loss calculated on the verification set Val , and the calculation formula of the loss function value Loss Val is as follows: ; Wherein mse() is a mean square error calculation function, P 6R,Val is the measured total pressure conversion value after the low-pressure turbine in the verification set, is the low-pressure turbine total pressure conversion value predicted value output by the model, F R,Val is the measured thrust parameter conversion value in the verification set, is the thrust parameter conversion value predicted value output by the model.

[0044] The present application measures the deviation of the low-pressure turbine total pressure conversion value predicted value F output by the model and the thrust parameter conversion value predicted value F by the mean square error, which not only pays attention to F R,Val directly related to the wing thrust, but also takes into account the P 6R,ValAs a check, avoid the performance of the model caused by single parameter measurement misjudgment, provide clear quantitative standard for model selection of one-time feature migration. Solve the problem of subjective performance judgment in the prior art, ensure that the "performance optimal partial model" selected from w instance models is not only suitable for thrust prediction, but also suitable for the measured P 6R,Val deviation; the quantized loss index (Loss Val ) improves the objectivity and reliability of model selection, lays a good candidate foundation for subsequent "optimal test model selection combined with historical flight monitoring data" in one-time feature migration, and indirectly ensures the accuracy of in-flight thrust evaluation.

[0045] Further, in step 3, the converted ground reference data is divided into training set and validation set, and the neural network model is trained for multiple rounds to obtain multiple instance models. The specific process of selecting the partial instance model with the best performance on the validation set through the preset loss index is as follows: The converted ground reference data is divided into training set and validation set, and the neural network model is trained for w rounds. Each round of training generates an instance model, and w instance models are obtained. The loss function value Loss Val of the w instance models on the validation set is calculated, and the m instance models with the smallest loss function value Loss Val are selected, where 2≤m

[0046] Multiple rounds of training cover different initial parameters and model states of data batches, reducing the accidental deviation of single round models. Retaining m optimal models instead of one not only selects long and short term memory neural network instance models with good performance, but also retains model diversity, provides sufficient candidates for subsequent further adaptation combined with flight monitoring data, and solves the problem of insufficient generalization ability and weak anti-interference of single model in the prior art. The setting of m candidate models ensures that there are enough performance excellent models to choose from in one-time feature migration, avoids the failure of high-altitude scene caused by poor adaptability of single model, improves the adaptation potential of the model to flight data, and provides protection for selecting the optimal test model that truly adapts to high-altitude working conditions.

[0047] Further, in step 3, the converted historical flight monitoring data is input into the screened instance model as auxiliary domain data, and the deviation between the converted key parameter values of the turbine after the converted historical flight monitoring data output by each instance model and the measured key parameter values of the turbine after the historical flight monitoring data is compared. The specific process of selecting the instance model with the smallest deviation as the optimal test model is as follows: The converted historical flight monitoring data is input into the above m instance models, and the low-pressure turbine total pressure conversion value prediction value , the mean square error value between the converted low-pressure turbine total pressure conversion value P 6R,Hist and the measured low-pressure turbine total pressure conversion value P in the historical flight monitoring data is minimized as the optimal test model, that is

[0048] Further, the specific process of step 4 is as follows: Step 4.1: Constructing a linear relationship formula based on a linear regression algorithm: input the converted flight monitoring data to be evaluated into the optimal test model, and construct a linear relationship formula based on the optimal prediction value of the turbine key parameter conversion value output by the optimal test model and the measured turbine key parameter conversion value in the converted flight monitoring data to be evaluated, and the linear relationship formula is as follows: ; Wherein, P 6R,Real is the measured low-pressure turbine total pressure conversion value in the flight monitoring data to be evaluated, is the optimal prediction value of the turbine key parameter conversion value output by the optimal test model; h1, h2 are correction coefficients to be solved; δ is the error of linear transformation; Step 4.2: Defining the matrix and vector required to solve the correction coefficient: Suppose the number of flight monitoring data samples to be evaluated for calculating the correction coefficient is n, then: Input matrix Z: the measured low-pressure turbine total pressure conversion value P 6R,Real of n flight monitoring data samples to be evaluated and the constant 1 are respectively taken as two columns to form a matrix, and the matrix expression is: ; Coefficient vector H: composed of correction coefficients h1, h2 to be solved, and the vector expression is ; Output vector Y: composed of n flight monitoring data samples to be evaluated corresponding to , and the vector expression is: ; Error vector δ: composed of the linear transformation errors of n flight monitoring data samples to be evaluated, and the vector expression is ; Step 4.3: Deriving the correction coefficient solving formula and calculating the coefficient vector H: converting the linear relationship formula in step 4.1 into a matrix form Y=ZH+δ; In order to minimize the square sum of the error vector δ, the solving formula of the correction coefficient vector H is derived by the least square method: ; Wherein, Z T is the transpose matrix of the input matrix Z, which is obtained by interchanging the rows and columns of the input matrix Z; Z T Z is the transpose matrix Z TThe product matrix of the input matrix Z, calculated according to the matrix multiplication rule; (Z T Z) −1 is the inverse matrix of the matrix ZTZ, calculated by the matrix inversion operation; The input matrix Z defined in step 4.2 and the output vector Y are substituted into the above solving formula to calculate the coefficient vector H; Step 4.4: Extract the correction coefficient and calibrate the thrust parameter conversion value optimal prediction value: The coefficient vector H calculated in step 4.3 is , the first element is extracted as the correction coefficient h1, and the second element is extracted as the correction coefficient h2; The converted aero-engine flight monitoring data to be evaluated is input into the optimal test model to obtain the thrust parameter conversion value optimal prediction value output by the optimal test model; The thrust parameter conversion value optimal prediction value is calibrated according to the calibration formula to obtain the final aero-engine in-wing thrust conversion value final prediction value ; the calibration formula is: ; Wherein is the final aero-engine in-wing thrust conversion value final prediction value, is the thrust parameter conversion value optimal prediction value output by the optimal test model, and h1 and h2 are correction coefficients; The calculation formula of the engine physical thrust final prediction value according to the thrust parameter conversion formula is as follows: ; Wherein: is the engine physical thrust final prediction value, is the final aero-engine in-wing thrust conversion value final prediction value, and X F is the thrust correction coefficient, 101.325 is the standard environmental reference pressure, and P1 is the fan inlet total pressure.

[0049] Step 4 considers the similarity of the ground reference data and the historical flight monitoring data, the to-be-evaluated flight monitoring data generation mechanism, that is, the feature distribution among the three types of data is similar, but there is scaling and offset, and the correction coefficients h1 and h2 are calculated based on the optimal prediction value of the turbine key parameter conversion value output by the optimal test model and the turbine key parameter conversion value in the to-be-evaluated flight monitoring data. The turbine key parameter conversion value and the engine thrust parameter conversion value have strong correlation, so h1 and h2 can be used to correct the optimal prediction value of the thrust parameter conversion value output by the optimal test model. The entire process does not require additional hardware, and only data and algorithms are used to realize bias compensation, solve the problem of large actual deviation of the thrust result in the prior art, and finally convert the calibrated in-flight aircraft engine thrust conversion value final prediction value into the engine physical thrust final prediction value, realizing accurate quantitative evaluation of the in-flight aircraft engine thrust.

[0050] Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can modify the technical solutions recorded in the foregoing embodiments or make equivalent replacements to some of the technical features, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An aeroengine in flight thrust assessment method, characterised in that: The method comprises the following steps: Step 1: data acquisition and equivalent conversion: acquiring ground reference data, historical flight monitoring data and to-be-evaluated flight monitoring data of an aero-engine, wherein the ground reference data comprises state parameters representing the running state of the engine and a thrust parameter as an evaluation target, and the historical flight monitoring data and the to-be-evaluated flight monitoring data comprise state parameters representing the running state of the engine; the ground reference data, the historical flight monitoring data and the to-be-evaluated flight monitoring data are converted to standard environmental conditions according to preset rules, to obtain converted ground reference data, historical flight monitoring data and to-be-evaluated flight monitoring data; Step 2: data-driven model construction: constructing a neural network model according to the structure of the engine, taking the state parameters in the converted ground reference data as the input of the model, and taking the converted values of the key parameters after the turbine and the converted values of the thrust parameters as the output of the model; Step 3: first feature migration: dividing the converted ground reference data into a training set and a validation set, training the neural network model for multiple rounds to obtain multiple instance models, and selecting part of the instance models with the best performance on the validation set through a preset loss index; inputting the converted historical flight monitoring data into the selected instance models as auxiliary domain data, comparing the deviations of the converted values of the key parameters after the turbine output by each instance model based on the converted historical flight monitoring data and the converted values of the key parameters after the turbine measured in the historical flight monitoring data, and selecting the instance model with the smallest deviation as an optimal test model; Step 4: second feature migration: inputting the converted to-be-evaluated flight monitoring data into the optimal test model to obtain the optimal prediction values of the converted values of the thrust parameters and the converted values of the key parameters after the turbine output by the optimal test model; calculating a correction coefficient based on a linear regression algorithm, using the optimal prediction values of the converted values of the key parameters after the turbine output by the optimal test model and the converted values of the key parameters after the turbine measured in the converted to-be-evaluated flight monitoring data; calibrating the optimal prediction values of the converted values of the thrust parameters through the correction coefficient to obtain the final prediction values of the converted values of the in-flight thrust of the aero-engine; and converting the final prediction values of the converted values of the in-flight thrust of the aero-engine into the final prediction values of the physical thrust of the engine based on a thrust conversion formula.

2. The aeroengine in-flight thrust assessment method of claim 1, wherein: The state parameters in step 1 comprise control parameters, gas path section parameters and variable geometry parameters; The control parameters comprise a high-pressure rotor speed N1, a low-pressure rotor speed N2 and a throttle lever angle PLA; The gas path cross-sectional parameters include the total temperature at the fan inlet (T1), the total pressure at the fan inlet (P1), and the total pressure at the compressor outlet (P). 31 Low-pressure turbine after total pressure P6, low-pressure turbine after total temperature T6; The variable geometry parameters comprise a fan guide vane angle a1, a compressor guide vane angle a2 and an engine tail nozzle angle D8; The thrust parameter is an engine physical thrust F; The key parameter after the turbine is a low-pressure turbine rear total pressure P6 or a low-pressure turbine rear total temperature T6.

3. The aeroengine in-flight thrust assessment method of claim 2, wherein: The ground reference data and the flight monitoring data are converted to standard environmental conditions according to preset rules, wherein the specific conversion process of high-pressure rotor speed N1, low-pressure rotor speed N2, compressor outlet total pressure P 31 , low-pressure turbine rear total pressure P6, low-pressure turbine rear total temperature T6, and engine physical thrust F is as follows: The high-pressure rotor speed conversion formula is as follows: ； Where: N 1R is the converted high-pressure rotor speed, N1 is the measured high-pressure rotor speed, X N1 is the high-pressure rotor speed correction factor, 288.15 is the standard ambient reference temperature, T1 is the measured fan inlet total temperature, and 273.15 is the temperature conversion factor; The low-pressure rotor speed conversion formula is as follows: ； Where: N 2R is the converted low-pressure rotor speed, N2 is the measured low-pressure rotor speed, X N2 is the low-pressure rotor speed correction factor, 288.15 is the standard ambient reference temperature, T1 is the measured fan inlet total temperature, and 273.15 is the temperature conversion factor; The compressor outlet total pressure conversion formula is as follows: ； where: P 31R is the corrected compressor exit total pressure, P 31 is the measured compressor exit total pressure, 101.325 is the standard ambient reference pressure, and P1 is the measured fan inlet total pressure; The low-pressure turbine rear total pressure conversion formula is as follows: ； where: P 6R P6 is the measured low pressure turbine discharge total pressure, 101.325 is the standard ambient reference pressure, and P1 is the measured fan inlet total pressure; and the low pressure turbine discharge total temperature conversion formula is as follows: ； Wherein: T 6R is the converted low-pressure turbine total temperature, T6 is the measured low-pressure turbine total temperature, X T6 is the low-pressure turbine total temperature correction coefficient, and 273.15 is the temperature conversion coefficient. The thrust parameter conversion formula is as follows: ； Where: F R is the scaled thrust parameter, F is the measured engine physical thrust, X F is the thrust correction factor, 101.325 is the standard ambient reference pressure, P1 is the measured fan inlet total pressure.

4. The aeroengine in-flight thrust assessment method of claim 1, wherein: The neural network model in step 2 is a long short-term memory neural network model, which includes a parameter input module, a feature extraction module, a feature coupling module, a feature dimension reduction module, and a result output module; the working process of each module is as follows: The parameter input module concatenates l pieces of continuous time series input parameter data into a single input sample, where l is the number of continuous time series data; The feature extraction module extracts features from the input sample and outputs a feature vector with a size of [l, c1], where c1 is a preset hidden feature dimension; The feature coupling module uses a bidirectional long short-term memory neural network model to perform bidirectional information coupling on the feature vector, and outputs a coupled feature vector with a length of [l, c2], where c2 is a preset hidden feature dimension of the bidirectional long short-term memory neural network model; The feature dimension reduction module uses a fully connected neural network to perform flattening and dimension reduction processing on the coupled feature vector; The result output module outputs a two-dimensional result ​ including the model output low pressure turbine after total pressure corrected value prediction and the model output thrust parameter corrected value prediction .

5. The aeroengine in-flight thrust assessment method of claim 1, wherein: In step 3: the preset loss index is a loss function value Loss calculated on the validation set Val The calculation formula of the loss function value Loss Val is as follows: ； where mse() is the mean square error calculation function, P 6R,Val is the measured low-pressure turbine after total pressure conversion value of the validation set, is the predicted value of the low-pressure turbine after total pressure conversion value output by the model, F R,Val is the measured thrust parameter conversion value of the validation set, is the predicted value of the thrust parameter conversion value output by the model.

6. The aeroengine in flight thrust assessment method of claim 5, wherein: In step 3, based on the converted ground reference data, a training set and a validation set are divided, and the neural network model is trained for multiple rounds to obtain multiple instance models. The specific process of selecting the part of instance models with the best performance on the validation set through a preset loss indicator is as follows: The converted ground reference data is divided into a training set and a verification set, the neural network model is trained for w rounds, each round generates an instance model, and w instance models are obtained; the loss function value Loss of the w instance models on the verification set is calculated Val , and m instance models with the smallest loss function value Loss Val are selected, where 2≤m<w.

7. The aeroengine on wing thrust assessment method of claim 6, wherein: In step 3, the converted historical flight monitoring data is input into the screened instance model as auxiliary domain data, and the deviation between the turbine key parameter conversion value output by each instance model based on the converted historical flight monitoring data and the measured turbine key parameter conversion value in the historical flight monitoring data is compared. The specific process of selecting the instance model with the smallest deviation as the optimal test model is as follows: The converted historical flight monitoring data is input into the m instance models, and a low-pressure turbine total pressure conversion value prediction value based on the converted historical flight monitoring data is obtained from each instance model The instance model with the minimum mean square error value between the calculated low-pressure turbine total pressure conversion value P 6R,Hist and the measured low-pressure turbine total pressure conversion value P in the historical flight monitoring data is taken as the optimal test model.

8. The aeroengine over-the-wing thrust assessment method of claim 1, wherein: The specific process of step 4 is as follows: Step 4.1: Construct a linear relationship formula based on a linear regression algorithm: Input the converted to-be-evaluated flight monitoring data into the optimal test model, and construct a linear relationship formula based on the turbine key parameter conversion value optimal prediction value output by the optimal test model and the measured turbine key parameter conversion value in the converted to-be-evaluated flight monitoring data. The linear relationship formula is as follows: ； wherein P 6R,Real is a measured low-pressure turbine rear total pressure conversion value in the flight monitoring data to be evaluated, is an optimal prediction value of the turbine rear key parameter conversion value output by the optimal test model; h1 and h2 are correction coefficients to be solved; and δ is an error of linear transformation. Step 4.2: Define the matrix and vector required to solve the correction coefficient: Let the number of to-be-evaluated flight monitoring data samples used to calculate the correction coefficient be n, then: Input matrix Z: measured low pressure turbine after total pressure corrected values P of n flight monitoring data samples to be evaluated 6R,Real With constant 1 as two columns respectively, the matrix expression is: ; The coefficient vector H is composed of the correction coefficients h1, h2 to be solved, and the vector expression is ; Output vector Y: corresponding to n flight monitoring data samples to be evaluated The vector expression is: ; Error vector δ: composed of the linear transformation errors of n flight monitoring data samples to be evaluated, vector expression is ; Step 4.3: Derive the correction coefficient solving formula and calculate the coefficient vector H: Convert the linear relationship formula in step 4.1 to matrix form Y=ZH+δ; To minimize the sum of squares of the error vector δ, the solution formula of the correction coefficient vector H is derived by the least square method: ; wherein Z T is the transpose matrix of the input matrix Z, obtained by interchanging the rows and columns of the input matrix Z; Z T Z is the transpose matrix Z T is the product matrix of the input matrix Z and the transpose matrix Z, calculated according to the matrix multiplication rule;(Z T Z) −1 is the inverse matrix of the matrix ZTZ, calculated by the matrix inversion operation; Substitute the input matrix Z and output vector Y defined in step 4.2 into the above solving formula to calculate the coefficient vector H; Step 4.4: Extract the correction coefficient and calibrate the thrust parameter conversion value optimal prediction value: coefficient vector calculated from step 4.3 In this case, the first element is extracted as the correction coefficient h1 and the second element is extracted as the correction coefficient h2. The converted to-be-evaluated aero-engine flight monitoring data is input into the optimal test model, and a converted thrust parameter conversion value optimal prediction value output by the optimal test model is obtained ; a calibrated formula for converting the thrust parameter converted value into an optimal predicted value performing calibration to obtain a final predicted value of the in-flight engine thrust converted value ; the calibrated formula is: ； wherein is the final prediction of the in-flight engine thrust conversion value, is the optimal prediction of the thrust parameter conversion value output by the optimal test model, and h1 and h2 are correction coefficients. According to the thrust parameter conversion formula, the calculation formula of the final prediction value of the engine physical thrust is as follows: ； wherein: is the final predicted value of the engine physical thrust, is the final predicted value of the aircraft engine in wing thrust conversion, X F is the thrust correction coefficient, 101.325 is the standard ambient reference pressure, and P1 is the fan inlet total pressure.

Images