A hypersonic aircraft engine inlet aerodynamic instability early warning method

Abstract

The invention discloses a method for early warning of aerodynamic instability of an inlet of a hypersonic aircraft engine. After normalizing the pre-processed wall pressure data characteristics, a binary classification of start-up and non-start-up state data is performed based on a linear SVM algorithm, an optimal classification surface is determined as a non-start-up boundary of the current state of the inlet, and an η% stability margin is defined to determine its non-start-up warning boundary. The non-start-up boundary is used to realize instability judgment of the inlet. On this basis, the back pressure multiple attribute of the wall pressure data along the way is trained and evaluated based on a BP neural network, and the wall pressure along the way at the non-start-up boundary and its warning boundary is used as a regression input. The non-start-up state of the inlet and the back pressure multiple boundary value of its warning state are predicted through a BP neural network model, which can be used to realize aerodynamic instability early warning of a hypersonic aircraft engine within a specific Mach number and flight altitude range. A new feasible scheme is provided for early warning of aerodynamic instability of the inlet of a hypersonic aircraft engine.

Description

Technical Field

[0001] The present invention relates to the field of aerodynamic instability of a hypersonic air inlet, and in particular to a method for early warning of aerodynamic instability of an air inlet of a hypersonic aircraft engine. Background Art

[0002] As an important component of a hypersonic aircraft engine, the air inlet plays a vital role in ensuring the working efficiency, normal operation, thrust size, etc. of the engine. The current aerodynamic instability problem of the hypersonic air inlet is mainly attributed to the factors of too high back pressure in the combustion chamber or too low Mach number of the incoming flow and too large angle of attack, which will lead to the occurrence of air inlet actuation. Since the actuation state of the hypersonic air inlet cannot be solved in the air intake design process, the current measures taken are to avoid the actuation of the air inlet or to restart the air inlet. Therefore, establishing a reasonable early warning mechanism based on the instability judgment for the actuation factor has important research significance and engineering application value for ensuring the stable flight of the hypersonic thrust system.

[0003] Regarding the problem of hypersonic inlet failure to start, the excessive back pressure in the combustion chamber is an important factor that causes aerodynamic instability (failure to start) in the hypersonic inlet. It is characterized by the appearance of a positive shock wave downstream of the inlet, which will be pushed out of the inlet under the excitation of high back pressure, resulting in a decrease in channel flow, thereby causing the inlet to fail to start. From the perspective of aerodynamic instability, based on a certain hypersonic inlet model, numerical simulation is performed under different back pressure multiples to obtain the wall pressure data along the inlet channel, and the learning algorithm is used to predict the instability warning boundary. Therefore, the method of hypersonic inlet instability warning presents some characteristics: First, in data processing, the amount of wall pressure data in the aerodynamic instability state of the inlet is small, resulting in an unbalanced distribution of stable state data and unstable state data, and the samples are high-dimensional and small-scale. However, accurate classification can be obtained through feature selection algorithm and SVM algorithm. The instability discrimination model with high accuracy and strong generalization ability can accurately classify the wall pressure instability data, thereby determining the classification boundary of the inlet duct starting state and non-starting state; second, from the perspective of the back pressure multiple, since a higher back pressure multiple will form a higher back pressure causing the inlet duct to not start, the back pressure multiple boundary of the inlet duct aerodynamic instability can be delineated between the existing starting state data and the non-starting state data through the regression method, and a certain stability margin can be set to obtain the back pressure multiple boundary of the inlet duct aerodynamic instability warning as a warning condition for the inlet duct aerodynamic instability. Summary of the invention

[0004] The purpose of the present invention is to provide a hypersonic inlet aerodynamic instability warning method based on machine learning to address the problem of inlet aerodynamic instability. It is currently a novel aerodynamic instability warning method.

[0005] The present application proposes a method for early warning of aerodynamic instability in the inlet of a hypersonic aircraft engine, which is characterized in that after normalizing the pre-processed along-the-way wall pressure data characteristics, a binary classification of the start-up and non-start-up state data is performed based on a linear SVM algorithm, an optimal classification surface is determined as the non-start-up boundary of the current state of the inlet, and an η% stability margin is defined to determine its non-start-up warning boundary, and the non-start-up boundary is used to realize the inlet instability judgment. On this basis, the back-pressure multiple attribute of the along-the-way wall pressure data is trained and evaluated based on a BP neural network, and the along-the-way wall pressure at the non-start-up boundary and its warning boundary is used as regression input. The non-start-up state of the inlet and the back-pressure multiple boundary value of its warning state are predicted by the BP neural network model, which can be used to realize the aerodynamic instability early warning of a hypersonic aircraft engine within a specific Mach number and flight altitude range.

[0006] In some specific embodiments, the method for early warning of aerodynamic instability of an aircraft engine inlet is characterized by comprising the following steps:

[0007] Step 1: wall pressure data collection and preprocessing;

[0008] Step 2: Normalize the input samples;

[0009] Step 3: Extract two-dimensional features of wall pressure;

[0010] Step 4: Calculate the non-activation boundary B of the two-dimensional feature classification of the wall pressure NS and does not activate the warning boundary B NSW ;

[0011] Step 5: Regression training and index evaluation of the back pressure multiple of the wall pressure;

[0012] Step 6: Estimation of the back pressure multiples in the non-starting state and the warning state;

[0013] Step 7: The instability warning conditions are achieved.

[0014] In some specific embodiments, the aerodynamic instability warning method for an aircraft engine inlet is characterized by a specific implementation method of step 1: setting pressure signal sampling points for a combined inlet model of a hypersonic aircraft engine model, each modal channel is divided into a number of along-the-way walls, and 100 pressure signal collection points are set on each along-the-way wall. At a typical operating point of the combined engine, a set of back pressure multiples is set to simulate the actual combustion chamber back pressure conditions, thereby obtaining the current pressure sample data of the inlet starting and not starting.

[0015] In some specific embodiments, the aerodynamic instability warning method for the aircraft engine inlet is characterized by the specific implementation method of step 4: using the SVM classification algorithm to perform linear binary classification on the start-up and non-start-up states of the wall pressure, determine the optimal classification surface, that is, the non-start-up boundary, which can be used as a detection basis for the inlet not starting in the current state, and define the η% stability margin to complete the determination of the non-start-up warning boundary of the inlet.

[0016] In some specific embodiments, the aerodynamic instability warning method for an aircraft engine inlet is characterized by the specific implementation method of step 6: for a sample set and its back pressure conditions, a BP neural network is used to perform regression training and evaluation feedback on the back pressure multiple of the inlet wall pressure along the way, and then based on the non-start boundary and the wall pressure along the warning boundary determined in step 3 as regression input, the BPNN regression model is used to predict the inlet non-start back pressure multiple boundary value and the η% warning condition back pressure multiple boundary value.

[0017] In some specific embodiments, the aerodynamic instability warning method for an aircraft engine inlet is characterized by a specific implementation method of step 7: based on the estimated values ​​of the back pressure multiple boundaries in the non-starting state and the warning state of a typical operating point, without considering the mode conversion, a rough instability boundary and a warning boundary curve are fitted within the flight Mach number and altitude range as a condition for warning of aerodynamic instability in the inlet on the flight track.

[0018] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0019] The present invention sets a back pressure multiple for a typical working point on a flight envelope of a combined cycle engine from the perspective of aerodynamic instability (non-starting) of an inlet caused by excessive back pressure in a combustion chamber, so as to simulate the back pressure condition of a hypersonic inlet, and obtains the wall pressure data along the channel through computer simulation. A support vector machine (SVM) algorithm is used to perform binary classification of starting state data and non-starting state data, and the classification accuracy reaches 100.00%, and the non-starting boundary of each working point is obtained, the instability judgment of the non-starting state data is completed, and the η% stability margin is defined to determine the non-starting warning boundary of each working point. On this basis, the wall pressure along the way is regressed as a back pressure multiple value, and a BP neural network is used for training and evaluation. Through a regression model with a learning rate α adjusted, the pressure data at the non-starting boundary and its warning boundary are used as input to predict the back pressure multiple boundary value of the inlet non-starting state and its warning state. Finally, the instability boundary and warning boundary values ​​predicted by the working points under each mode (turbine, ejector rocket and ramjet) are fitted to obtain the non-starting boundary and warning boundary of inlet aerodynamic instability within a specific Mach number and flight altitude range without considering the mode conversion, which are used as the warning conditions for inlet aerodynamic instability.

[0020] Compared with other traditional studies on the problem of early warning of aerodynamic instability in the air intake, the present invention considers the situation where the air intake fails to start due to excessive back pressure in the combustion chamber of the thrust system, and uses machine learning methods to complete the instability boundary judgment and early warning scheme implementation for the wall pressure data along the air intake channel and the corresponding starting state and back pressure multiple conditions.

[0021] The present invention provides the distribution of aerodynamic instability boundaries and warning boundaries of a hypersonic aircraft engine within a flight track. For turbine, ejector rocket and ramjet modes, by predicting the back pressure multiple instability boundaries and warning boundary values ​​of typical working points under each mode, the thrust system aerodynamic instability and warning boundary curves are fitted within the flight track (without considering the mode conversion stage), which is used to provide instability warning for aerodynamic instability of the inlet on the flight track. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 Flow chart of the air inlet aerodynamic instability warning method in an embodiment of the present invention;

[0023] Figure 2 The wall pressure distribution at different channel positions along a certain working point in the embodiment of the present invention;

[0024] Figure 3 It is the SVM binary classification of the non-starting and starting states of a certain working point in the embodiment of the present invention;

[0025] Figure 4 It is the schematic diagram of the BP neural network structure used;

[0026] Figure 5 Schematic diagram of the distribution of the inlet instability boundary and its warning boundary curve within a flight track. DETAILED DESCRIPTION

[0027] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention clearer and more understandable, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0028] The present embodiment is a method for early warning of aerodynamic instability in the inlet of a hypersonic aircraft engine. By collecting pressure data under conditions of low back pressure multiples to high back pressure multiples set along the wall of the inlet, after preprocessing and data normalization, two pressure features are extracted to complete the SVM binary classification of the wall pressure start-up and non-start-up states. At the same time, the BP neural network is used to complete the regression of the back pressure multiple of the wall pressure, and the wall pressure at the non-start boundary of the SVM is used as input to regress the non-start state back pressure multiple boundary value, and the η% stability margin is set to obtain the non-start state warning back pressure multiple boundary value. Finally, within the range of flight Mach number and altitude, the aerodynamic instability boundary curve and warning curve of the inlet on the flight track are fitted, which can be used to perform early warning processing of the aerodynamic instability of the inlet on the flight track.

[0029] The following combination Figure 1 The implementation method of the present invention is further described.

[0030] Step 1: Wall pressure data collection and preprocessing.

[0031] 1-1. First, set pressure signal sampling points for the combined inlet model of the hypersonic aircraft engine. Each modal channel is divided into several along-the-way walls, and 100 pressure signal collection points are set on each along-the-way wall. At the typical working point of the combined engine, a set of back pressure multiples is set to simulate the actual combustion chamber back pressure conditions, and then obtain the current pressure sample data of the inlet start and non-start (a single back pressure multiple condition can simulate several wall pressure samples with 100-dimensional features).

[0032] 1-2. Process outliers and missing values ​​of the pressure sample data collected by numerical simulation and merge and sort them. Figure 2 The wall pressure distribution along the different channel positions (features) of a certain working point is shown. Each pressure distribution contains 100 data points, where the red dotted line represents the wall pressure data in the non-starting state obtained by simulating the non-starting back pressure multiple condition, and the blue solid line represents the wall pressure data in the starting state obtained by simulating the starting back pressure multiple condition.

[0033] Step 2: Normalize the input samples.

[0034] For input sample T = {(x k ,y k )|x k ∈R n ,y k ∈R n ,k=1,2,…,n}, and record x k (i) (i=1,2,…,m) is the kth sample data corresponding to the i-th feature, where y krepresents the label of the kth sample, m represents the number of features of the input sample (100 dimensions), and n represents the number of input samples. Use the following formula:

[0035]

[0036] The input samples are normalized to (0,1) so that the values ​​of each feature sample are in the same order of magnitude. k (i) (max) and x k (i) (min) They represent the maximum and minimum values ​​of the sample data corresponding to the i-th feature, respectively, and x k (i) ' represents the sample data after normalization of the i-th feature.

[0037] Step 3: Extract two-dimensional features of wall pressure.

[0038] For the normalized input sample T'={(x k ',y k )|x k '∈R n ,y k ∈R n , k=1,2,…,n}, and the chi-square test feature selection algorithm is used to select the two-dimensional features of the wall pressure {(x (p) ,x (q) )|p,q=1,2,…,100,p≠q}.

[0039] Step 4: Calculate the non-activation boundary B of the two-dimensional feature classification of the wall pressure NS and does not activate the warning boundary B NSW .

[0040] 4-1. Based on the extracted two-dimensional features of the wall pressure, the input samples are randomly divided into 80% training set and 20% test set, and the SVM linear separable binary classification of the start and non-start of each working point is completed, and the optimal classification surface can be obtained:

[0041]

[0042] x (p) and x (q) The two characteristic variables representing the wall pressure, where ω = [ω0, ω1] and b are the hyperplane parameters of the SVM linearly separable, so the boundary B is not activated. NS The equation can be expressed as ω0x (p) +ω1x (q) +b=0. Figure 3It shows that the SVM linearly separable two-classification (reaching 100% classification accuracy) of the wall pressure along the working point in the non-starting and starting states.

[0043] 4-2. Based on step 4-1, define η% stability margin and determine the warning boundary B with η% stability margin for the non-start boundary. NSW , its equation can be expressed as ω0x (p) +ω1x (q) +b w =0, where b w =(1-η%)·b.

[0044] Step 5: Regression training and index evaluation of the back pressure multiple of the wall pressure.

[0045] 5-1. For the wall pressure data, the given back pressure multiple is used as the data label, and the input sample is normalized to (0,1) by the method of step 2, and then the data label is also normalized to obtain the normalized regression input sample T r '={(x k ',y (r)k ')|x k '∈R n ,y (r)k ∈R n ,k=1,2,…,n}.

[0046] 5-2. The two-dimensional features {x (p) ,x (q)}, the wall pressure sample data corresponding to this two-dimensional feature is T r (p,q) '={(x k (i) ',y (r)k ')|x k (i) '∈R n×2 ,y (r)k ∈R n ,i={p,q},k=1,2,…,n}, divide it into 80% regression training set and 20% regression test set, and use BP neural network for training and evaluation.

[0047] 5-3, Figure 4 The network structure of BP neural network is shown. Given BP neural network input (x1, x2) = {x k (p) ',x k (q) '|k=1,2,…,n},output y={y (r)k')|k=1,2,…,n}. Set 2 hidden layers, the weight and bias coefficient of the connection between the input layer and the hidden layer 1 are {(w i,1 [1] ,w i,2 [1] ,b i [1] )|i=1,2,…,10}, the weight and bias coefficient of the connection between hidden layer 1 and hidden layer 2 are {(w i,1 [2] ,w i,2 [2] ,…,w i,10 [2] ,b i [2] )|i=1,2,…,20}, ReLU is used as the activation function; the weight and bias coefficient of the connection between hidden layer 2 and the output layer are {(w 1,1 [3] ,w 1,2 [3] ,…,w 1,20 [3] ,b1 [3] )}, using a linear function.

[0048] 5-4. Taking a typical operating point in a flight envelope of a hypersonic combined cycle engine as the object, the wall pressure sample data is trained and predicted using the BPNN with initialized parameters. Finally, the prediction results are denormalized and output, and then the coefficient R is determined. 2 The root mean square error is used as the evaluation index of regression training. The learning rate α is further adjusted according to the index results to improve it and determine the appropriate α parameter value of the network.

[0049] Step 6: Estimation of the back pressure multiples in the non-starting state and the warning state.

[0050] 6-1. Based on steps 4 and 5, the wall pressure data at the non-activated boundary is used as the regression input for the non-activated back pressure multiple prediction, which is specifically determined by the non-activated boundary equation ω0x (p) +ω1x (q) +b=0 and the least squares equation of the sample distribution in the two-dimensional feature space after normalization x (q) =p0x (p) +p1 is determined. Its input is The non-starting back pressure multiple boundary value M is obtained through BPNN prediction BP (ω,b).

[0051] 6-2. From the boundary equation ω0x without starting the warning (p) +ω1x (q) +b w= 0, and the least squares equation of the normalized two-dimensional feature space sample distribution can determine the warning boundary input as where b w =(1-η%)·b, and the non-starting warning back pressure multiple boundary value M is obtained through BPNN prediction BP (ω,b w ).

[0052] Step 7: The instability warning conditions are achieved.

[0053] By estimating the back pressure multiple boundaries of the non-starting state and the warning state of the typical working point, without considering the mode conversion, for the single-mode (turbine, ejector rocket and ramjet mode) thrust system working stage, the approximate instability boundary and warning boundary curve are fitted within a specific Mach number and flight altitude range as the condition for the inlet aerodynamic instability warning within the flight Mach number and altitude range. Figure 5 The approximate distribution of the hypersonic inlet non-starting back pressure multiple boundary curve S1 and its η% warning boundary curve S2 within the flight Mach number and altitude range is shown.

[0054] The combustion chamber back pressure multiplier value when the hypersonic inlet works at the current state point (Ma, H) is M BP At this time, the values ​​corresponding to curve S1 and curve S2 are S1 and * and S2 * , then the method for realizing the inlet instability warning without starting can be expressed in the following form:

[0055]

[0056] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

[0057] The above describes the specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A hypersonic aircraft engine inlet aerodynamic instability early warning method, characterized in that: The steps include: Step 1: wall pressure sample data collection and preprocessing, specifically: setting pressure signal sampling points along the wall of the combined inlet model of the hypersonic aircraft engine model, setting the back pressure multiple at the typical working point of the combined engine to obtain the wall pressure sample data of the current inlet starting and not starting; Step 2: input the wall pressure sample data for normalization; Step 3, extracting the two-dimensional features of the wall pressure sample data; Step 4: Calculate the inactivation boundary B of the two-dimensional feature classification of the wall pressure sample data NS and does not activate the warning boundary B NSW Specifically, the starting and non-starting state data are classified into two categories based on the linear SVM algorithm, the optimal classification surface is determined as the non-starting boundary of the current state of the intake duct, and the η% stability margin is defined to determine the non-starting warning boundary, and the non-starting boundary is used to realize the intake duct instability judgment; Step 5: Back pressure multiple regression training and index evaluation of the wall pressure sample data. On this basis, the back pressure multiple attribute of the wall pressure sample data is trained and evaluated based on the BP neural network; Step 6, estimation of the back pressure multiples in the non-start state and the warning state, specifically: using the wall pressure sample data at the non-start boundary and its warning boundary as regression input, and predicting the back pressure multiple boundary values ​​of the inlet non-start state and its warning state through the BP neural network model; Step 7, the instability warning condition is realized, specifically: according to the estimated value of the back pressure multiple boundary of the non-starting state and the warning state of the typical working point, without considering the mode conversion, the instability boundary and the warning boundary curve are fitted within the flight Mach number and altitude range, which are used as the condition of the aerodynamic instability warning of the air inlet on the flight track, so as to realize the aerodynamic instability warning of the hypersonic aircraft engine within the specific Mach number and flight altitude range.

2. The method for early warning of aerodynamic instability of an aircraft engine inlet according to claim 1, characterized in that The specific implementation method of step 1 is as follows: pressure signal sampling points are set for the combined inlet model of the hypersonic aircraft engine model, each modal channel is divided into a number of along-the-way walls, 100 pressure signal collection points are set on each along-the-way wall, and at the typical working point of the combined engine, a set of back pressure multiples is set to simulate the actual combustion chamber back pressure conditions, thereby obtaining the wall pressure sample data of the current inlet start-up and non-start-up.

Images