Aero-engine performance monitoring method and system based on cross-section parameter adaptive evaluation and medium

Abstract

The present application belongs to the technical field of aero-engine performance monitoring, and discloses an aero-engine performance monitoring method, system and medium based on cross-section parameter adaptive evaluation. The aero-engine performance monitoring method based on cross-section parameter adaptive evaluation comprises an aero-engine simulation data set generation method, a Monte Carlo training method of an aero-engine component-level network and a flow network, a Monte Carlo training method of an aero-engine cross-section parameter adaptive model, a flight data retraining method of the aero-engine cross-section parameter adaptive model, and a deployment method of the aero-engine cross-section parameter adaptive model. Compared with a traditional cross-section parameter threshold monitoring method, the model and algorithm of the present application can calculate the cross-section parameter value under normal performance degradation according to different flight conditions and operating points of the engine, more reasonably evaluate the performance degradation state of the engine, reduce the false alarm rate and the missed alarm rate of the cross-section parameter, and improve the accuracy of the aero-engine monitoring process.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of aero-engine performance monitoring, and particularly relates to an aero-engine performance monitoring method and system for cross-section parameter adaptive evaluation and a medium. BACKGROUND

[0002] At present, aero-engine performance monitoring methods are mainly divided into two categories: a physical model-based method and a data-driven method.

[0003] The physical model-based method first establishes a thermodynamic model of an aero-engine, introduces a performance degradation factor in a component characteristic map of the model, and optimizes the degradation factor by means of an optimization method such as a gradient algorithm, a genetic algorithm, or a Kalman filtering method, with the objective of reducing measurement errors, so as to evaluate the performance degradation of each component. However, due to the interference of factors such as model assumption errors, component characteristic map errors, engine assembly errors, and sensor measurement errors, the error of the engine physical model is difficult to control, which further affects the evaluation accuracy of the engine degradation state. Therefore, the physical model-based method has gradually been replaced by the data-driven method.

[0004] The data-driven aero-engine performance monitoring method reduces the dependence of the model on the component characteristic map, avoids the time-consuming iteration process of the thermodynamic model, and therefore has faster efficiency and improved accuracy, becoming a research hotspot for aero-engine performance monitoring at home and abroad. In actual field work, a cross-section parameter such as exhaust temperature is often used to monitor the performance of an aero-engine.

[0005] Through the above analysis, the problems and defects of the prior art are as follows:

[0006] The cross-section parameter monitoring method widely used in the prior art for aero-engines is still limited to a simple threshold method, that is, corresponding thresholds are set for each state of the engine, and when the cross-section parameter of the engine in a certain state exceeds the threshold, the performance state of the aero-engine is warned. However, this method has the following shortcomings:

[0007] (1) The cross-section parameter threshold method does not consider the influence of engine flight conditions and working states on the cross-section parameter. The engine cross-section parameter is influenced by many factors, and at present, only a rough division of the engine states (such as slow, cycle, intermediate, small boost, maximum, etc.) is made, that is, only the influence of different working states on the engine cross-section parameter is simply considered. The cross-section parameter threshold method does not consider the influence of engine flight conditions such as ambient temperature, Mach number, and atmospheric pressure, nor does it consider the influence of a single engine speed or fuel flow on the cross-section parameter, that is, it does not give the standard value of the cross-section parameter of the engine at a certain working point.

[0008] (2) Given the cross-section parameter threshold is relatively strict, resulting in false alarm too much. In order to ensure flight safety, in the absence of accurate cross-section parameter standard value, the cross-section parameter threshold is usually given relatively strict, resulting in cross-section parameter is easy to exceed due to external environmental changes (such as temperature rise, inhale tail gas). However, the engine performance does not appear abnormal at this time, resulting in excessive maintenance, increasing the economic cost and time cost.

[0009] (3) It is difficult to monitor the performance degradation trend of the aero-engine. Since only the cross-section parameter threshold is given, this method can only monitor the current working state of the aero-engine, and it is difficult to comprehensively evaluate the performance degradation trend, which leads to the difficulty in timely discovering the cross-section parameter drift caused by the early performance failure of the aero-engine, and easily causes safety accidents.

[0010] The difficulty of solving the above problems and defects is that the cross-section parameter evaluation is the key to solving the above problems, and the influence of the engine flight condition and working process on the cross-section parameter is taken into account in the calculation process of the cross-section parameter to obtain a cross-section parameter calculation model. However, the accuracy of the engine performance evaluation depends on the modeling accuracy of the cross-section parameter model, and the traditional thermodynamic modeling method has low modeling freedom, which is difficult to comprehensively cover the entire working envelope and degradation process of the engine, and cannot achieve adaptive adjustment of the cross-section parameter.

[0011] The significance of solving the above problems and defects is that through adaptive evaluation of the cross-section parameter of the aero-engine, the influence of the external conditions and degradation state of the aero-engine on the cross-section parameter is fully considered, which is an effective means of performance monitoring of the aero-engine, and has important guiding significance for maintenance work in the field and flight standards of the aircraft.

[0012] In view of the problems of the cross-section parameter threshold, the difficulties of cross-section parameter modeling and the significance of cross-section parameter evaluation for performance monitoring of the aero-engine, the present application provides a cross-section parameter adaptive evaluation method for performance monitoring of the aero-engine, which fully considers the influence factors of the engine flight condition and working point on the cross-section parameter, and gives a reasonable cross-section parameter adaptive evaluation method, which can monitor the cross-section parameter of the aero-engine in real time, and can also monitor the performance degradation of the aero-engine through the change trend of the cross-section parameter. SUMMARY

[0013] To overcome the problems in the related art, the present application provides a cross-section parameter adaptive evaluation method for performance monitoring of the aero-engine, system and medium.

[0014] The technical solution is as follows: a cross-section parameter adaptive evaluation method for performance monitoring of the aero-engine, comprising:

[0015] The thermodynamic model of the aero-engine is combined, and the neural network model is used to obtain the variation law of the aero-engine cross-section parameters with flight conditions and working points; and the variation law obtained is used as a monitoring model for actual use of the aero-engine to monitor the performance state of the aero-engine.

[0016] In an embodiment, the performance state of the engine is monitored in real time by acquiring the monitoring model or on-board.

[0017] In an embodiment, the aero-engine performance monitoring method based on adaptive evaluation of cross-section parameters specifically comprises:

[0018] S1, generation of aero-engine simulation data set: the probability distribution of flight conditions and working points is extracted from aero-engine flight data, and a Monte Carlo method is used to construct an aero-engine simulation data set by using an aero-engine thermodynamic model;

[0019] S2, Monte Carlo training of aero-engine component-level network and flow network: the simulation data set generated in S1 is used to train the aero-engine component-level network and flow network;

[0020] S3, Monte Carlo training of aero-engine cross-section parameter adaptive model: the aero-engine flow network trained in S2 is connected to the aero-engine component-level network in S1 to replace the difficult-to-measure air flow input in the component-level network, forming an aero-engine cross-section parameter adaptive model, and the model is pre-trained using the data set in S1;

[0021] S4, flight data retraining of aero-engine cross-section parameter adaptive model: the input and output data required by the adaptive model established in S3 are screened from flight data and processed to form a flight data set, and the aero-engine cross-section parameter adaptive model is retrained;

[0022] S5, deployment of aero-engine cross-section parameter adaptive model: by comparing the cross-section parameter error distribution diagrams of the cross-section parameter adaptive model on the standard flight data and the to-be-tested flight data, a conclusion is given on whether the performance and degradation trend of the aero-engine is normal.

[0023] In an embodiment, the step S1 specifically comprises:

[0024] S11, a thermodynamic model is constructed based on the component characteristic map of a certain type of aero-engine, and the thermodynamic model is corrected by using actual test or flight data, and the final aero-engine model is represented by the following formula:

[0025] {T i , P i , W i}=engine(T2,P2,Pamb , N2, map), where i is the section number, T i , P i , W i are the total temperature, total pressure and flow rate of the i-th section, respectively, P amb is the ambient pressure, and map is the map of the engine;

[0026] S12, construct the joint probability distribution of the total temperature, total pressure, ambient pressure and high pressure speed of the engine inlet using the flight data:

[0027] T2, P2, P amb , N2~ F(t, p, p amb , n2),

[0028] Based on the above probability distribution, 5120 simulation input data are generated using the Monte Carlo method:

[0029] T 2j , P 2j , P ambj , N 2j ~ F(t, p, p amb , n2),

[0030] where the subscript j represents the j-th sample subject to the joint probability density F(t, p, p amb , n2), and the samples are collected to form the simulation input data set:

[0031] Ω input = {T 2j , P 2j , P ambj , N 2j}, j = 1, 2, …, 5120;

[0032] S13, input the above data set into the thermodynamic model of the aero-engine to obtain the parameters of each section of the aero-engine:

[0033] {T ij , P ij , W ij , Wf j , N 1j} = engine(T 2j , P 2j , P ambj , N 2j , map), where the subscript ij represents the i-th section parameter calculated by the j-th input sample, Wf j is the fuel flow rate calculated by the j-th input sample, and N 1j is the low pressure speed calculated by the j-th input sample;

[0034] S14, integrate the simulation input data set generated by S12 and the cross-section parameters calculated by the aero-engine thermodynamic model in S13 into an aero-engine simulation data set, and divide the data set into a training set and a validation set according to a ratio of 0.8:0.2:

[0035] Ω train ={T ij , P ij , W ij , P ambj , Wf j , N 1j , N 2j}, j = 1, 2, …, 4096,

[0036] Ω val ={T ij , P ij , W ij , P ambj , Wf j , N 1j , N 2j}, j = 1, 2, …, 1024.

[0037] In an embodiment, the step S2 specifically comprises the following steps:

[0038] S21, an aero-engine component-level model is constructed by using a neural network structure, wherein the input parameters of the aero-engine component-level model include: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed, engine air flow, and internal air flow; and the output parameters are aero-engine cross-section parameters, including cross-section total temperature, total pressure and flow.

[0039] S22, the neural network weights are optimized by using the simulation training set generated by S14, and the weights with the minimum error of the simulation validation set in the optimization process are selected as the final optimized component-level network weights.

[0040] S23, an aero-engine flow model is constructed by using a neural network structure, wherein the input parameters of the aero-engine flow model include: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed; and the output parameters include engine air flow and internal air flow.

[0041] S24, the neural network weights are optimized by using the simulation training set generated by S14, and the weights with the minimum error of the simulation validation set in the optimization process are selected as the final optimized component-level network weights.

[0042] In an embodiment, the step S3 specifically comprises the following steps:

[0043] S31, based on the component-level network model obtained in S22, inputting the output of the flow network model obtained in S24 to complete the construction process of the aero-engine cross-section parameter adaptive model; the input parameters of the constructed aero-engine cross-section parameter adaptive model include: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed; the output parameters include total temperature, total pressure and flow of each cross-section;

[0044] S32, using the simulation training set generated in S14 to optimize the aero-engine cross-section parameter adaptive model constructed in S31, and selecting the weight with the minimum error in the simulation verification set in S14 during the optimization process as the weight of the finally optimized cross-section parameter adaptive model.

[0045] In an embodiment, the step S4 comprises the following steps:

[0046] S41, extracting engine-related data from the flight data in the whole life cycle of performance degradation, which should at least include: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed and at least one cross-section parameter; taking the ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed as input and the cross-section parameter as output to form a flight data set; selecting the early data (such as the first 10% data of the whole life cycle) as the new machine standard training set and the late data as the standard performance degradation set;

[0047] S42, using the above new machine standard training set, retraining the adaptive model established in S32 by using the SGD optimization method or the Adam optimization method, and selecting the model with the smallest difference (i.e. the highest accuracy) between the predicted value and the true value of the cross-section parameter contained in the flight data as the final aero-engine cross-section parameter adaptive model.

[0048] In an embodiment, the step S5 comprises the following steps:

[0049] S51, using the cross-section parameter adaptive model established in S42 to predict the cross-section parameter of the standard performance degradation set in S41, and using the absolute error or relative error calculation method to calculate the error between the predicted value and the actual value in the test set, and drawing a standard error trend chart of the error changing with the engine use time or cycle number;

[0050] S52, after each flight, the data required by S41 is extracted from the flight data, the cross-section parameter prediction value is calculated by using the cross-section parameter adaptive model established by S42, the test error of the cross-section parameter actual value in the flight data is calculated by using the same error calculation method, the test error is compared with the upper limit of the standard error distribution of the similar use time or cycle number, if all the test errors are lower than the upper limit of the standard error distribution, the cross-section parameter margin is normal, otherwise the cross-section parameter margin is abnormal;

[0051] S53, the data required by S41 is extracted from the flight data in the recent period of time, the cross-section parameter prediction value is calculated by using the cross-section parameter adaptive model established by S42, the error of the cross-section parameter actual value in the flight data is calculated by using the same error calculation method, and the test error distribution graph is drawn; the test error distribution graph is compared with the standard error distribution graph of the same use time or cycle number, whether the above two error distribution graphs belong to the same distribution is tested by using t test and KS test method, if yes, the conclusion information that the trend of the cross-section parameter of the aero-engine is normal is output, otherwise, the conclusion information that the trend of the cross-section parameter is abnormal is output.

[0052] Another purpose of the application is to provide a cross-section parameter adaptive evaluation aero-engine performance monitoring system for implementing the cross-section parameter adaptive evaluation aero-engine performance monitoring method, comprising:

[0053] An aero-engine simulation data set module is used for generating an aero-engine simulation data set: the probability distribution of flight conditions and working points is extracted from aero-engine flight data, and the aero-engine simulation data set is constructed by using the aero-engine thermodynamic model by using the Monte Carlo method;

[0054] An aero-engine component-level network and flow network training module is used for Monte Carlo training of an aero-engine component-level network and flow network: the simulation data set generated by the aero-engine simulation data set module is used to train the aero-engine component-level network and flow network;

[0055] An aero-engine cross-section parameter adaptive model pre-training module is used for Monte Carlo training of an aero-engine cross-section parameter adaptive model: the aero-engine flow network trained by the aero-engine component-level network and flow network training module is connected to the aero-engine component-level network of the aero-engine simulation data set module to replace the difficult-to-measure air flow input in the component-level network, form an aero-engine cross-section parameter adaptive model, and pre-train the model by using the data set of the aero-engine simulation data set module;

[0056] The aero-engine section parameter adaptive model retraining module is used for flight data retraining of the aero-engine section parameter adaptive model, filtering input and output data required by the adaptive model established by the aero-engine section parameter adaptive model pretraining module from flight data, processing the input and output data, forming a flight data set, and retraining the aero-engine section parameter adaptive model;

[0057] The aero-engine performance and degradation trend acquisition module is used for deployment of the aero-engine section parameter adaptive model, comparing section parameter error distribution maps of the section parameter adaptive model on standard flight data and to-be-tested flight data, and giving a conclusion on whether the aero-engine performance and degradation trend are normal.

[0058] Another object of the present application is to provide a receiving user input program storage medium, and the stored computer program enables the electronic device to perform the aero-engine performance monitoring method for section parameter adaptive evaluation.

[0059] In combination with all the above technical solutions, the present application has the following advantages and positive effects:

[0060] The present application provides an aero-engine performance monitoring method for section parameter adaptive evaluation, which comprises: an aero-engine simulation data set generation method; a Monte Carlo training method for aero-engine component-level networks and flow networks; a Monte Carlo training method for an aero-engine section parameter adaptive model; a flight data retraining method for the aero-engine section parameter adaptive model; and a deployment method for the aero-engine section parameter adaptive model.

[0061] Experimental results and analysis show that, compared with the traditional section parameter threshold monitoring method, the model and algorithm of the present application can calculate the section parameter values under normal performance degradation according to different flight conditions and working points of the engine, more reasonably evaluate the performance degradation state of the engine, reduce the false alarm rate and the missed alarm rate of the section parameter, and improve the accuracy of the aero-engine monitoring process

[0062] It is understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the disclosure of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0063] The accompanying drawings incorporated in the specification and constituting a part hereof illustrate embodiments consistent with the present disclosure and, together with the specification, serve to explain principles of the present disclosure.

[0064] Figure 1 FIG. 1 is a flowchart of an aero-engine performance monitoring method for section parameter adaptive evaluation provided by an embodiment of the present application.

[0065] Figure 2is a full connection neural network-based aero-engine component-level model schematic diagram of step S21 of at least one embodiment provided by the embodiment of the present application.

[0066] Figure 3 is a full connection neural network-based aero-engine flow model schematic diagram of step S23 of at least one embodiment provided by the embodiment of the present application.

[0067] Figure 4 is an aero-engine cross-section parameter adaptive model schematic diagram of step S32 of at least one embodiment provided by the embodiment of the present application.

[0068] Figure 5 is a standard exhaust temperature error trend graph with flight time of step S51 of a certain type of aero-engine standard performance degradation set of 0 to 730 hours of at least one embodiment provided by the embodiment of the present application.

[0069] Figure 6 is a test exhaust temperature error trend graph with flight time of step S52 of the 328th hour of at least one embodiment provided by the embodiment of the present application.

[0070] Figure 7 is a test exhaust temperature error trend graph with flight time of step S53 of the 296th to 306th hour of at least one embodiment provided by the embodiment of the present application.

[0071] Figure 8 is a standard exhaust temperature error frequency distribution graph and test exhaust temperature error frequency distribution graph comparison graph of step S53 of at least one embodiment provided by the embodiment of the present application.

[0072] Among them, Figure 8(a) is a standard exhaust temperature error frequency distribution graph; Figure 8(b) is a test exhaust temperature error frequency distribution graph comparison graph. DETAILED DESCRIPTION

[0073] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below in conjunction with the drawings. In the following description, a lot of specific details are set forth in order to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the spirit of the present application, therefore the present application is not limited to the specific embodiments disclosed below.

[0074] In view of the problems of excessive false alarms and difficulty in finding early faults in the current aero-engine performance monitoring method based on simple cross-section parameter threshold, the application provides an aero-engine performance monitoring method based on adaptive evaluation of cross-section parameters.

[0075] Specifically, the application provides an aero-engine performance monitoring method based on adaptive evaluation of cross-section parameters, which comprises the following steps:

[0076] S1, generation of an aero-engine simulation data set: the probability distribution of flight conditions and working points is extracted from aero-engine flight data, and a Monte Carlo method is used to construct an aero-engine simulation data set by using an aero-engine thermodynamic model.

[0077] S2, Monte Carlo training of an aero-engine component-level network and a flow network: the simulation data set generated in S1 is used to train the aero-engine component-level network and the flow network.

[0078] S3, Monte Carlo training of an aero-engine cross-section parameter adaptive model: the aero-engine flow network trained in S2 is connected to the aero-engine component-level network in S1 to replace the difficult-to-measure air flow input in the component-level network, thereby forming an aero-engine cross-section parameter adaptive model, and the model is pre-trained by using the data set in S1.

[0079] S4, flight data retraining of the aero-engine cross-section parameter adaptive model: the input and output data required by the adaptive model established in S3 are screened from flight data and processed to form a flight data set, and the aero-engine cross-section parameter adaptive model is retrained.

[0080] S5, deployment of the aero-engine cross-section parameter adaptive model: the conclusion on whether the performance and degradation trend of the aero-engine are normal is given by comparing the cross-section parameter error distribution diagrams of the cross-section parameter adaptive model on the standard flight data and the to-be-tested flight data.

[0081] The application further provides an aero-engine performance monitoring system based on adaptive evaluation of cross-section parameters, which comprises:

[0082] An aero-engine simulation data set module is used to generate an aero-engine simulation data set: the probability distribution of flight conditions and working points is extracted from aero-engine flight data, and a Monte Carlo method is used to construct an aero-engine simulation data set by using an aero-engine thermodynamic model.

[0083] The aero-engine component-level network and flow network training module is used for Monte Carlo training of aero-engine component-level networks and flow networks: it uses the simulation dataset generated by the aero-engine simulation dataset module to train the aero-engine component-level networks and flow networks.

[0084] The pre-training module for the adaptive model of aero-engine cross-section parameters is used for Monte Carlo training of the adaptive model of aero-engine cross-section parameters: The aero-engine flow network trained by the aero-engine component-level network and flow network training module is connected to the aero-engine component-level network of the aero-engine simulation dataset module, replacing the air flow input that is difficult to measure in the component-level network, forming the adaptive model of aero-engine cross-section parameters, and the model is pre-trained using the dataset of the aero-engine simulation dataset module.

[0085] The retraining module for the adaptive model of aero-engine cross-section parameters is used for retraining the adaptive model of aero-engine cross-section parameters using flight data. It selects and processes the input and output data required for the adaptive model established by the pre-training module of the adaptive model of aero-engine cross-section parameters from the flight data to form a flight dataset, and then retrains the adaptive model of aero-engine cross-section parameters.

[0086] The aero-engine performance and degradation trend acquisition module is used for the deployment of the aero-engine cross-section parameter adaptive model: by comparing the cross-section parameter error distribution map of the cross-section parameter adaptive model on standard flight data and test flight data, it gives a conclusion on whether the aero-engine performance and degradation trend are normal.

[0087] The technical solution of the present invention will be further described below with reference to the embodiments and accompanying drawings.

[0088] Example

[0089] like Figure 1 As shown, an embodiment of the present invention provides a method for monitoring the performance of aero-engines with adaptive evaluation of cross-sectional parameters, comprising the following steps:

[0090] S1. Extract the probability distribution of flight conditions and operating points from the flight data of a certain type of aero-engine, and construct an aero-engine simulation dataset using the Monte Carlo method and the aero-engine thermodynamic model.

[0091] S2 uses the simulation dataset generated by S1 to train the component-level network and flow network of the aero-engine.

[0092] S3, access the aero-engine flow network trained in S2 to the aero-engine component-level network in S1, replace the air flow input difficult to measure in the component-level network, form an aero-engine cross-section parameter adaptive model, and pre-train the model by using the data set in S1.

[0093] S4, screen the input and output data required by the adaptive model established in S3 from the flight data, process the flight data set, and re-train the aero-engine cross-section parameter adaptive model.

[0094] S5, give a conclusion on whether the aero-engine performance and degradation trend are normal by comparing the cross-section parameter error distribution diagrams of the cross-section parameter adaptive model on the normal flight data and the to-be-tested flight data.

[0095] According to at least one embodiment of the present application, step S1 comprises the following steps:

[0096] S11: construct a thermodynamic model based on the component characteristic map of a certain type of aero-engine, and correct the thermodynamic model by using actual test or flight data, and finally the aero-engine model can be represented by the following formula:

[0097] {T i , P i , W i} = engine (T2, P2, P amb , N2, map),

[0098] In the formula, i is the cross-section number, T i , P i , W i are the total temperature, total pressure and flow of the i-th cross-section respectively, P amb is the ambient pressure, and map is the component characteristic map of the engine;

[0099] S12: construct the joint probability distribution of the inlet total temperature, total pressure, ambient pressure and high-pressure rotating speed of the aero-engine by using flight data:

[0100] T2, P2, P amb , N2 ~ F (t, p, p amb , n2),

[0101] Based on the above probability distribution, 5120 simulation input data are generated by using the Monte Carlo method:

[0102] T 2j , P 2j , P ambj , N 2j ~ F (t, p, p amb , n2),

[0103] where subscript j denotes the jth sample subject to the joint probability density F(t, p, p amb , n2), and the samples are collected to form a simulation input dataset:

[0104] Ωi nput = {T 2j , P 2j , P ambj , N 2j}, j = 1, 2, …, 5120;

[0105] S13: input the above dataset into the aero-engine thermodynamic model to obtain aero-engine section parameters:

[0106] {T ij , P ij , W ij , Wf j , N 1j} = engine(T 2j , P 2j , P ambj , N 2j , map),

[0107] where subscript ij denotes the ith section parameter calculated by the jth input sample, Wf j is the fuel flow calculated by the jth input sample, and N 1j is the low-pressure rotating speed calculated by the jth input sample;

[0108] S14: integrate the simulation input dataset generated by S12 and the section parameters calculated by the aero-engine thermodynamic model in S13 into an aero-engine simulation dataset, and divide the dataset into a training set and a validation set according to a ratio of 0.8:0.2:

[0109] Ω train = {T ij , P ij , W ij , P ambj , W fj , N 1j , N 2j}, j = 1, 2, …, 4096,

[0110] Ω val = {T ij , P ij , W ij , P ambj , Wf j , N 1j , N 2j}, j = 1, 2, …, 1024;

[0111] According to at least one embodiment disclosed in the present application, step S2 comprises the following steps:

[0112] S21: An aero-engine component-level network model is constructed using the full connection neural network structure as shown in Figure 2 The model input includes: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed, engine air flow, and the output is aero-engine cross-section parameters, including total temperature, total pressure and flow rate of each cross-section.

[0113] S22: The Adam optimization method is adopted, the step size is 0.001, the iteration number is 300, the loss function is mean square error loss, the simulation training set generated by S14 is used to optimize the neural network weight, and the weight with the minimum error of the simulation verification set in the optimization process is selected as the final optimized weight of the component-level network.

[0114] S23: An aero-engine flow network is constructed using the full connection neural network structure as shown in Figure 3 The model input includes: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed, and the output is engine air flow and internal air flow.

[0115] S24: The Adam optimization method is adopted, the step size is 0.001, the iteration number is 300, the loss function is mean square error loss, the simulation training set generated by S14 is used to optimize the neural network weight, and the weight with the minimum error of the simulation verification set in the optimization process is selected as the final optimized weight of the flow network.

[0116] According to at least one embodiment disclosed in the present application, step S3 comprises the following steps:

[0117] S31: As shown in Figure 4 Based on the component-level network model obtained by S22, the input is replaced by the output of the flow network model obtained by S24, and the construction process of the aero-engine cross-section parameter adaptive model is completed. The model input after construction is: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed, and the output includes total temperature, total pressure and flow rate of each cross-section.

[0118] S32: The Adam optimization method is adopted, the step size is 0.001, the iteration number is 300, the loss function is mean square error loss, the simulation training set generated by S14 is used to optimize the neural network weight, and the weight with the minimum error of the simulation verification set in the optimization process is selected as the final optimized weight of the cross-section parameter adaptive model.

[0119] According to at least one embodiment of the present disclosure, step S4 comprises the following steps:

[0120] S41: Extract engine-related data from the flight data of the 730-hour normally-performing aero-engine of the type, including: ambient static pressure, engine inlet total temperature and total pressure, fuel flow, high and low pressure rotor speed, and discharge temperature. Take the ambient static pressure, engine inlet total temperature and total pressure, fuel flow, high and low pressure rotor speed as inputs, and the discharge temperature as output to form a flight data set containing 14786 data. Take the first 1500 as a new machine standard training set, and the last 13286 as a standard performance degradation set.

[0121] S42: Using the above new machine standard training set, using the Adam optimization method, with 1e-4 as the optimization step, 500 iterations, and the optimization step decaying by 20% every 100 steps, retraining the cross-section parameter adaptive model established in S32, with the mean square error of the model predicted discharge temperature value and the actual discharge temperature value in the data set as the optimization target.

[0122] According to at least one embodiment of the present disclosure, step S5 comprises the following steps:

[0123] S51: Using the cross-section parameter adaptive model established in S42 to predict the discharge temperature of the standard performance degradation set in S41, calculate the relative absolute error between the predicted discharge temperature and the actual value in the test set, and draw a standard error trend chart, as shown in Figure 5 .

[0124] S52: Extract 294 engine steady-state data from the flight data of the 328-hour flight, including: ambient static pressure, engine inlet total temperature and total pressure, fuel flow, high and low pressure rotor speed, and discharge temperature. Take the ambient static pressure, engine inlet total temperature and total pressure, fuel flow, high and low pressure rotor speed as inputs, and the discharge temperature as output to form a test data set. Calculate the predicted discharge temperature value using the cross-section parameter adaptive model established in S42, and calculate the error between the predicted discharge temperature value and the actual discharge temperature value in the flight data using the same error calculation method as S51, as shown in Figure 6 , the upper limit of the test discharge temperature error is 0.014; from the standard error trend chart established in S51, the upper limit of the standard discharge temperature error from 300 hours to 350 hours is 0.021, it can be seen that the upper limit of the test discharge temperature error is lower than the upper limit of the standard discharge temperature error, and a conclusion of normal discharge temperature margin is given.

[0125] S53: Extract 1526 steady engine data from flight data of flight 296-306 hours, including: ambient static pressure, engine inlet total temperature, total pressure, fuel flow, high and low pressure rotor speed and discharge temperature. The ambient static pressure, engine inlet total temperature, total pressure, fuel flow, high and low pressure rotor speed are taken as input, and the discharge temperature is taken as output to form the test data set. The cross-section parameter adaptive model established in S42 is used to calculate the predicted value of the discharge temperature, and the same error calculation method as S51 is used to calculate the error between the predicted value and the actual measured value of the discharge temperature in the flight data, as shown in Figure 7

[0126] Calculate the test error distribution histogram (Fig. 8(a)), and compare it with the error distribution histogram of hours 296-306 in the standard error trend chart (Fig. 8(b)). After K-S test, it is concluded that the two error distributions belong to the same distribution, so it is concluded that the discharge temperature trend of the aero-engine at this time point is normal.

[0127] Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses or adaptive changes of this disclosure that follow the general principles of the disclosure and include known equivalents or technical

[0128] It should be understood that the present disclosure is not limited to the precise structures as herein described and illustrated in the drawings, and that various modifications and changes can be made without departing from its scope. The scope of the present disclosure should be gauged by the appended claims and their legal equivalents.​

Claims

1. An aeroengine performance monitoring method with cross-section parameter adaptive assessment, characterized in that, The aero-engine performance monitoring method based on cross-section parameter adaptive evaluation comprises: The aero-engine performance monitoring method based on cross-section parameter adaptive evaluation comprises: The aero-engine performance monitoring method based on cross-section parameter adaptive evaluation comprises: S1, aero-engine simulation data set generation: the probability distribution of flight conditions and working points is extracted from aero-engine flight data, and a Monte Carlo method is used to construct an aero-engine simulation data set by using an aero-engine thermodynamic model; S2, Monte Carlo training of aero-engine component-level network and flow network: the simulation data set generated in S1 is used to train the aero-engine component-level network and flow network; S3, Monte Carlo training of aero-engine cross-section parameter adaptive model: the aero-engine flow network trained in S2 is connected to the aero-engine component-level network in S2 to replace the difficult-to-measure air flow input in the component-level network, forming an aero-engine cross-section parameter adaptive model, and the data set in S1 is used to pre-train the model; S4, flight data retraining of aero-engine cross-section parameter adaptive model: the input and output data required by the adaptive model established in S3 are screened from the flight data and processed to form a flight data set, and the aero-engine cross-section parameter adaptive model is retrained; S5, deployment of aero-engine cross-section parameter adaptive model: by comparing the cross-section parameter error distribution diagram of the cross-section parameter adaptive model on the standard flight data and the to-be-tested flight data, a conclusion is given on whether the performance and degradation trend of the aero-engine is normal.

2. The aeroengine performance monitoring method of claim 1, wherein, The monitoring model is obtained and used for real-time monitoring of the performance state of the engine.

3. The method of claim 1, wherein, The step S1 specifically comprises: S11, based on the component characteristic map of a certain type of aero-engine, a thermodynamic model is constructed, and the thermodynamic model is corrected by using actual test or flight data, and the final aero-engine model is represented by the following formula: {T i , P i , W i} = engine(T2, P2, P amb , N2, map), where i is the section number, T i , P i , W i are the total temperature, total pressure and mass flow rate of the i-th section, P amb is the ambient pressure, and map is the map of the engine components. S12, the joint probability distribution of the inlet total temperature, total pressure, ambient pressure and high-pressure rotating speed of the aero-engine is constructed by using flight data: T2, P2, P amb , N2~F(T, P, P amb , N2), Based on the above probability distribution, 5120 simulation input data are generated by using the Monte Carlo method: T 2j , P 2j , P ambj , N 2j ~ F(T, P, P amb , N2), where subscript j denotes the jth sample drawn from the joint probability density F(T, P, P amb , N2) and the samples are collected to form a simulated input dataset: Ω input = {T 2j , P 2j , P ambj , N 2j}, j = 1, 2,..., 5120; S13, input the above data set into the aero-engine thermodynamic model to obtain each cross-section parameter of the aero-engine: {T ij , P ij , W ij , Wf j , N 1j} = engine(T 2j , P 2j , P ambj , N 2j , map), where subscript ij represents the i-th cross-section parameter calculated from the j-th input sample, Wf j fuel flow calculated for the j-th input sample, N 1j low pressure speed of rotation calculated for the j-th input sample; S14, integrate the simulation input data set generated in S12 and each cross-section parameter calculated by the aero-engine thermodynamic model in S13 into an aero-engine simulation data set, and divide the data set into a training set and a validation set according to the proportions of 0.8 and 0.2: Ω train = {T ij , P ij , W ij , P ambj , Wf j , N 1j , N 2j}, j = 1, 2,..., 4096, Ω val = {T ij , P ij , W ij , P ambj , Wf j , N 1j , N 2j}, j = 1, 2,..., 1024.

4. The aeroengine performance monitoring method of claim 3, wherein, The step S2 specifically comprises the following steps: S21, a neural network structure is adopted to construct an aero-engine component-level model, wherein input parameters of the aero-engine component-level model include: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed, engine air flow and internal air flow; and output parameters of the aero-engine component-level model include aero-engine cross-section parameters, including cross-section total temperature, total pressure and flow; S22, the neural network weight is optimized by using the simulation training set generated in S14, and the weight with the minimum error of the simulation verification set in the optimization process is selected as the weight of the finally optimized component-level network; S23, a neural network structure is adopted to construct an aero-engine flow model, wherein input parameters of the aero-engine flow model include: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed; and output parameters of the aero-engine flow model include engine air flow and internal air flow; S24, the neural network weight is optimized by using the simulation training set generated in S14, and the weight with the minimum error of the simulation verification set in the optimization process is selected as the weight of the finally optimized component-level network.

5. The method of claim 4, wherein, The step S3 specifically includes the following steps: S31, based on the component-level network model obtained in S22, the input is changed to the output of the flow network model obtained in S24, and the construction process of the aero-engine cross-section parameter adaptive model is completed; Input parameters of the constructed aero-engine cross-section parameter adaptive model include: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed; and output parameters of the aero-engine cross-section parameter adaptive model include cross-section total temperature, total pressure and flow; S32, the aero-engine cross-section parameter adaptive model constructed in S31 is optimized by using the simulation training set generated in S14, and the weight with the minimum error of the simulation verification set in the optimization process is selected as the weight of the finally optimized cross-section parameter adaptive model.

6. The method of claim 5, wherein, The step S4 includes the following steps: S41, engine related data is extracted from flight data in the whole life cycle of performance degradation, which should at least include: ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed and at least one cross-section parameter; ambient static pressure, engine inlet airflow total temperature, total pressure, fuel flow, high and low pressure rotor speed are taken as input, and the cross-section parameter is taken as output to form a flight data set; the early data in the whole life cycle is selected in a certain proportion as a new machine standard training set, and the late data is selected as a standard performance degradation set; S42, the adaptive model established in S32 is retrained by using the above new machine standard training set by a certain optimization method, the output of the adaptive model is only the cross-section parameter contained in the flight data, and the model with the highest accuracy is selected as the final aero-engine cross-section parameter adaptive model.

7. The method of claim 6, wherein, The step S5 includes the following steps: S51, the cross-section parameter adaptive model established in S42 is used to predict the cross-section parameter of the standard performance degradation set in S41, a certain error calculation method is used to calculate the error between the predicted value of the cross-section parameter and the actual value in the test set, and a standard error trend graph of the error changing with the engine use time or cycle number is drawn; S52, after each flight, the data required by S41 is extracted from the flight data, the cross-section parameter prediction value is calculated by using the cross-section parameter adaptive model established by S42, the test error between the cross-section parameter prediction value and the actual value in the flight data is calculated by using the same error calculation method; compare the test error with the upper limit of the standard error distribution of the similar use time or cycle number, if all the test errors are lower than the upper limit of the standard error distribution, the cross-section parameter margin is normal, otherwise the cross-section parameter margin is abnormal; S53, from the flight data S41 required data in the recent period of time, the cross-section parameter prediction value is calculated by using the cross-section parameter adaptive model established by S42, the error between the cross-section parameter prediction value and the actual value in the flight data is calculated by using the same error calculation method, and the test error distribution graph is drawn; compare the test error distribution graph with the standard error distribution graph of the same use time or cycle number, and use a certain hypothesis testing method to test whether the above two error distribution graphs belong to the same distribution, if they belong to the same distribution, output the conclusion information that the trend of the aero-engine cross-section parameter is normal, otherwise, output the conclusion information that the trend of the cross-section parameter is abnormal.

8. An aeroengine performance monitoring system for implementing the aeroengine performance monitoring method of adaptive cross-section parameter estimation according to any one of claims 1 to 7, characterized in that, The aero-engine performance monitoring system of the cross-section parameter adaptive evaluation comprises: An aero-engine simulation data set module for generating an aero-engine simulation data set: extracting the probability distribution of flight conditions and working points from aero-engine flight data, and constructing an aero-engine simulation data set by using an aero-engine thermodynamic model through a Monte Carlo method; An aero-engine component-level network and flow network training module for Monte Carlo training of an aero-engine component-level network and flow network: training the aero-engine component-level network and flow network by using the simulation data set generated by the aero-engine simulation data set module; An aero-engine cross-section parameter adaptive model pre-training module for Monte Carlo training of an aero-engine cross-section parameter adaptive model: connecting the aero-engine flow network trained by the aero-engine component-level network and flow network training module to the aero-engine component-level network of the aero-engine simulation data set module to replace the difficult-to-measure air flow input in the component-level network, forming an aero-engine cross-section parameter adaptive model, and pre-training the model by using the data set of the aero-engine simulation data set module; An aero-engine cross-section parameter adaptive model retraining module for flight data retraining of an aero-engine cross-section parameter adaptive model: screening and processing the input and output data required by the adaptive model established by the aero-engine cross-section parameter adaptive model pre-training module from the flight data to form a flight data set, and retraining the aero-engine cross-section parameter adaptive model; An aero-engine performance and degradation trend acquisition module for deployment of an aero-engine cross-section parameter adaptive model: giving a conclusion whether the aero-engine performance and degradation trend is normal by comparing the cross-section parameter error distribution graphs of the cross-section parameter adaptive model on the standard flight data and the to-be-tested flight data.

9. A receiving user input program storage medium, the stored computer program causing an electronic device to perform the aeroengine performance monitoring method of adaptively evaluating the cross-section parameter according to any one of claims 1 to 7.

Images