A method for collaborative design of overall control of hypersonic vehicle based on multi-fidelity data fusion

Abstract

The application discloses a kind of based on high supersonic aircraft overall control collaborative design method of multi-fidelity data fusion, by introducing bayesian regression technique to convert least square problem equation of eigenvalue orthogonal decomposition into regression process, for fusing aircraft high reliability and low reliability aerodynamic force data, improve the global model precision of aerodynamic characteristics in aircraft design process;On this basis, the overall design method is balanced to realize the design of aerodynamic stability and control design well, and has the ability to process the optimization design problem of multiple state points in flight envelope, provides a kind of quick and reliable design tool for high supersonic aircraft.

Description

Technical Field

[0001] This invention relates to the field of aircraft design technology, and in particular to a collaborative design method for the overall control of hypersonic aircraft based on multi-fidelity data fusion. Background Technology

[0002] Traditional aircraft design methods typically employ a distributed design approach, breaking down the design task into multiple design disciplines, each relatively independent of the others. This design approach is efficient and feasible when the coupling between disciplines is weak, or when their mutual influence is linear.

[0003] For hypersonic vehicles, the subsystems typically exhibit strong nonlinear coupling and high sensitivity to design variables. Traditional distributed design methods often yield unsatisfactory results and fail to meet performance and stability requirements. Therefore, modern vehicle design must take a multidisciplinary approach to achieve high-performance solutions and truly unlock the performance potential of such vehicles.

[0004] During the development of an aircraft, a large amount of aerodynamic data is required across the entire flight envelope for different flight conditions. Aircraft aerodynamic data typically comes from three sources: numerical calculations, wind tunnel tests, and flight tests. Numerical calculations are considered low-reliability aerodynamic data acquisition methods; while relatively easy to obtain, they suffer from lower accuracy and difficulty in convergence. High-reliability aerodynamic data, usually with higher accuracy and cost, generally comes from wind tunnel and flight tests, but the amount of data is insufficient to meet the aircraft design requirements.

[0005] Therefore, in order to try to solve the above problems, it is necessary to carry out multidisciplinary collaborative design research based on improving the accuracy of the aerodynamic characteristic model of the aircraft, so as to enhance the credibility of the aircraft design and its engineering application value. Summary of the Invention

[0006] The technical problem to be solved by this invention is to provide a hypersonic vehicle overall control collaborative design method based on multi-fidelity data fusion. This method performs data fusion modeling on high and low confidence datasets when the number of high confidence samples is limited, expands the scope of aerodynamic data fusion and improves the accuracy of data fusion. This data fusion method is then applied to the field of aerodynamic and control collaborative design of hypersonic vehicles to enhance the reliability of vehicle design and its engineering application value.

[0007] To address the aforementioned technical problems, this invention provides a collaborative design method for the overall control of hypersonic vehicles based on multi-fidelity data fusion, comprising the following steps:

[0008] Step 1: Initialize the aircraft's geometric design variables and constraints;

[0009] Step 2: Determine the baseline aerodynamic layout of the hypersonic vehicle and use the state-type function method to parametrically describe the vehicle's geometric configuration;

[0010] Step 3: Based on the above parametric geometry, high-fidelity datasets are obtained by engineering estimation method and CFD fluid calculation method respectively. Multi-fidelity data fusion is performed on the high-fidelity datasets based on the Bayesian extended intrinsic orthogonal decomposition method to obtain the aircraft aerodynamic characteristic prediction surrogate model.

[0011] Step 4: Establish a dynamic characteristic model of the aerospace vehicle, and use the aerospace vehicle aerodynamic stability and control co-design method based on the spatial criterion diagram, combined with the flight envelope, to determine whether the vehicle meets the lateral aerodynamic stability requirements under multiple state points.

[0012] Step 5: Under the premise of meeting the above stability requirements, introduce a control system and combine it with control law design. By adjusting the controller parameters, meet the closed-loop control aerodynamic stability criteria and non-minimum phase requirements, and finally output the aerodynamic layout design results of the aircraft that meet the requirements of aerodynamic stability and control co-design.

[0013] Preferably, in step 1, the design variables are the wing sweep angle and fuselage height geometric parameters of the hypersonic vehicle.

[0014] Preferably, in step 2, the hypersonic aircraft adopts a hypersonic aircraft with a large swept double delta wing and no horizontal tail aerodynamic layout. The fuselage cross section has a sharp side edge spine configuration, the two sides of the fuselage are double swept delta wings, a pair of elevators on both sides control the pitch channel, and a single vertical tail design controls the lateral channel of flight.

[0015] Preferably, in step 3, the intrinsic orthogonal decomposition data fusion method based on Bayesian extension specifically includes the following steps:

[0016] Step 31: Initialize the design variables of the multi-confidence aerodynamic model, design experiments for different confidence models, obtain the corresponding model responses, and obtain high and low confidence datasets;

[0017] Step 32: Based on the low-confidence data model, the intrinsic orthogonal decomposition mode matrix is ​​obtained by singular value decomposition.

[0018] Step 33: Search and determine the coordinates of the low-fidelity data point closest to the high-confidence data. Based on the least squares method and combined with the above-mentioned intrinsic orthogonal decomposition mode matrix, the intrinsic orthogonal decomposition basis coefficients can be obtained.

[0019] Step 34: Based on Bayesian theory, the least squares problem of eigenorthogonal decomposition is transformed into a regression problem. For the eigenorthogonal decomposition mode matrix and the corresponding high-fidelity data set, a corresponding Gaussian process regression model is constructed.

[0020] Step 35: By evaluating the predicted distribution of all rows of the intrinsic orthogonal decomposition mode matrix, output the mean and variance information of the data fusion solution under different input conditions.

[0021] Preferably, in step 31, initializing the design variables of the multi-confidence aerodynamic model, designing experiments for different confidence levels, obtaining the corresponding model responses, and obtaining high and low confidence datasets specifically includes the following steps:

[0022] Step 31a: Design variables as x = [x1, x2, ..., x d ]∈D, where D is the design space, d is the dimension of the design space, and satisfies R is the set of real numbers;

[0023] Step 32b: Experimental design refers to sampling different confidence models within the design space D to obtain high and low confidence data sets, where the low confidence data set Y:=[y 1 ,...,y n ]=[y(ξ1),...,y(ξ n Design variable ξ for low-reliability data i ∈D, i=1,...,n, where n is the number of low-confidence datasets; the high-confidence dataset t:=[t1,...,t s ]=[y′(e1),...,y′(e s High-confidence data design variable e i ∈D, i=1,...,s; s is the number of high-confidence datasets. The sampling method is one of the following: optimal Latin hypercube sampling, full factorial design, or orthogonal experimental design, or it is assumed that the sample points are given.

[0024] Preferably, in step 32, the intrinsic orthogonal decomposition mode matrix is ​​obtained by calculating singular value decomposition based on the low-confidence data model as follows:

[0025] For the aforementioned low-fidelity dataset Y∈R N×n Singular value decomposition yields:

[0026] Y=U∑V T

[0027] In the formula, U=[u 1 ,...,u N ]∈R N×N and V = [v 1 ,...,vn ]∈R n×n It is an orthogonal matrix, i.e., U T U=UU T =I N and V T V = VV T =I n , where ∑=diag(σ1,...,σ n )∈R N×n Includes singular values ​​σ1≥...≥σ in descending order n ≥0;

[0028] Assuming the rank of matrix Y is r = rank(Y), then only the first r singular values ​​are non-zero. The corresponding r left singular vectors, i.e., the first r columns of matrix U, constitute a set of singular values ​​derived from the dataset y1,...,y n The orthonormal basis {u1,...,u} of the space formed r}, that is, the intrinsic orthogonal decomposition basis U r .

[0029] Preferably, in step 33, the coordinates of the low-fidelity data point closest to the high-confidence data are searched and determined. Based on the least squares method and combined with the above-mentioned intrinsic orthogonal decomposition mode matrix, the intrinsic orthogonal decomposition basis coefficients can be obtained as follows:

[0030] A given vector t∈R of high-confidence data s It can be defined as y′∈R N There are only components in the middle. Given a vector, where s < N is the number of sampling points, j1,...,j s ∈{1,...,N}:

[0031]

[0032] For a matrix in Indicates the j-th i There are s standard basis vectors. The components of vector t are represented by the components of vector y′ by a nearest neighbor search of the coordinates of s sampling points in the computational grid;

[0033] The vector y′ is obtained by approximation in the eigenorthogonal decomposition subspace, and the eigenorthogonal decomposition basis coefficients can be found. Make

[0034]

[0035] In the formula, U r =[u 1 ,...,u r ]∈R N×rThis is the eigenorthogonal decomposition basis vector matrix, i.e., the first r columns of U. The basis coefficient vectors... The L2 error that minimizes the observations for vector y′ is defined by the least squares problem.

[0036]

[0037] Typically, X = P T U r ∈R s×r Having full column rank, the above equation has a unique solution:

[0038]

[0039] This basic coefficient vector Substituting into the approximate equation, we can obtain an estimate of the vector y′.

[0040] Preferably, in step 34, based on Bayesian theory, the least squares problem of eigenorthogonal decomposition is transformed into a regression problem. Specifically, for the eigenorthogonal decomposition mode matrix and the corresponding high-fidelity dataset, a corresponding Gaussian process regression model is constructed as follows:

[0041] Based on the Bayesian extension concept, the least squares problem equation of eigenorthogonal decomposition is defined as a regression process, and a mapping f is constructed: Directly use matrix U r The row is mapped to the vector y′

[0042]

[0043] A set of data pairs is obtained at a given sampling location:

[0044] {(x i ,t i |i=1,...,s}

[0045] in U represents the mode matrix of the eigenorthogonal decomposition. r The jth i OK, For the corresponding sampled response;

[0046] The input-output relationship of the above data set is then surrogate using Gaussian process regression:

[0047] Suppose f(x) is a Gaussian process with zero mean, for any two inputs x, x′∈R r Its covariance Cov[f(x),f(x′)] is given by the covariance function k(x,x′), which adopts a widely used class of covariance functions as shown below:

[0048] k(x,x′)=θ0·exp(-θ1||xx′|| 2 )+θ2x T x′

[0049] In the formula, θ0, θ1, and θ2 are hyperparameters, which are usually not defined in advance but determined from the data by maximizing the log-boundary likelihood function.

[0050]

[0051] So, for a new input variable x * ∈R r f * := f(x * The predicted distribution of ) is

[0052] E[f * ]=k(x * (K+σ) 2 I) -1 t

[0053] Var[f * ]=k(x * ,x * )-k(x * ) T (K+σ 2 I) -1 k(x * )

[0054] Where k(x) * :=(k(x * ,x i )) i＝1,..,s ∈R s and K:=(k(x i ,x j )) i,j＝1,..,s ∈R s×s .

[0055] Preferably, in step 35, the mean and variance information of the data fusion solution under different input conditions are output by evaluating the predicted distribution of all rows of the intrinsic orthogonal decomposition mode matrix. Specifically, this is achieved by evaluating the predicted distribution of all rows x of the mode matrix. * =(U r ) i The predicted distribution of i = 1, ..., N is used to obtain the data fusion prediction results based on the Bayesian extended intrinsic orthogonal decomposition method.

[0056] The fusion evaluation criteria include the root mean square error (RMSE) and the coefficient of determination (R²). 2 And the interval evaluation criterion PIC:

[0057]

[0058]

[0059]

[0060] Where, N V To verify the sample size, y i To verify the true values ​​of the sample points, To verify the predicted values ​​of the sample points, For y i The average value of q is any quantile between (0,1), where q = 0.025 and q = 0.975.

[0061] Preferably, in step 4, a dynamic characteristic model of the aerospace vehicle is established. Based on the aerospace vehicle aerodynamic stability and control co-design method using a spatial criterion diagram, and combined with the flight envelope, it is determined whether the vehicle meets the lateral aerodynamic stability requirements at multiple state points. Specifically:

[0062] The attitude dynamics model of the hypersonic vehicle is constructed as follows:

[0063]

[0064]

[0065] In the formula, (α,β,μ) represent the angle of attack, sideslip angle, and roll angle of the aircraft; (p,q,r) represent the roll rate, pitch rate, and yaw rate; L and Y represent the lift and side force acting on the aircraft, respectively; L, M, and N represent the roll moment, pitch moment, and yaw moment acting on the aircraft, respectively; m and V represent the mass and velocity of the aircraft, respectively; I x I z I y These are the moments of inertia of each axis; I zx For the rotational product of each axis.

[0066] State points are selected based on the flight envelope, and the model is linearized at each state point. Based on the principle of small disturbances, the attitude dynamics equations can be decoupled into longitudinal and lateral motions after linearization using the horizontal no-sideslip flight condition. Ignoring gravity, lift, and lateral forces, the above equations are decoupled into longitudinal and lateral equations. Stability analysis and control system design are based on the following equations:

[0067]

[0068]

[0069] In the formula, L′ α L′β L′ p L r ′、L q ′、 These are the derivatives of the rolling torque for each channel; N′ α 、N′ β 、N′ p N r ′、N q ′、 These are the derivatives of the yaw moment for each channel; M′ α M′ β , M q ′ represents the derivative of the pitch moment for each channel; δ e Elevator deflection angle; δ a This refers to the aileron deflection angle.

[0070] The operators in the equation are defined as follows:

[0071]

[0072] For hypersonic vehicles, the lateral channel is usually unstable, and the selection of lateral stability parameters and lateral control laws is particularly important.

[0073] By solving the characteristic equation |λI-A|, the lateral stability criterion can be written as the following equation:

[0074]

[0075] In the formula, C nβ C lβ These are two classic stability parameters.

[0076] Based on the spatial criterion graph method, the x-axis and y-axis are respectively mapped to the lateral stability parameter C. lβ and C nβ The above equation can be written as:

[0077]

[0078] This allows us to draw a spatial criterion diagram for lateral open-loop stability, determining whether the lateral stability requirements are met at multiple state points within the flight envelope of the aircraft under this configuration. If not, we select new geometric parameters and return to step 2; if they are met, we proceed to step 5.

[0079] Preferably, in step 5, under the premise of satisfying the above stability requirements, a control system is introduced in conjunction with the control law design. By adjusting the controller parameters, the closed-loop control aerodynamic stability criterion and the non-minimum phase requirements are satisfied. The final output is the aerodynamic layout design result of the aircraft that satisfies the requirements of aerodynamic stability and control coordinated design. Specifically:

[0080] Since the lateral path of the aircraft is unstable, a control law is required for stabilization. Therefore, the design of the control system is introduced, and the closed-loop control aerodynamic stability criterion is as follows:

[0081]

[0082] In the formula, H and F represent different closed-loop stability regions.

[0083] Similarly, this corresponds to the spatial criterion diagram:

[0084]

[0085] Furthermore, according to the transverse non-minimum phase criterion, there exists a straight line passing through the origin:

[0086]

[0087] In the formula, These are the aerodynamic stability parameters caused by the ailerons.

[0088] When the stability parameter (C) lβ C nβ When the point corresponding to the line is above the line, it indicates that the lateral system of the aircraft is a non-minimum phase system, and it can be controlled by conventional control methods.

[0089] The beneficial effects of this invention are as follows: By introducing Bayesian regression technology, this invention transforms the least squares problem equation of intrinsic orthogonal decomposition into a regression process, which is used to fuse high-confidence and low-confidence aerodynamic data of the aircraft, thereby improving the accuracy of the global aerodynamic characteristic model in the aircraft design process; on this basis, the overall design method achieves a good balance between aerodynamic stability design and control design, and has the ability to handle the optimization design problem of multiple state points within the flight envelope, providing a fast and reliable design tool for hypersonic aircraft. Attached Figure Description

[0090] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0091] Figure 2 This is a schematic diagram of the pneumatic data fusion method of the present invention.

[0092] Figure 3This is a schematic diagram of the collaborative design process for aircraft aerodynamic stability and control according to the present invention. Detailed Implementation

[0093] like Figure 1 As shown, a collaborative design method for the overall control of a hypersonic vehicle based on multi-fidelity data fusion includes the following steps:

[0094] Step 1: Initialize the aircraft's geometric design variables and constraints;

[0095] The design variables and constraints are defined according to the actual problem. For example, the design variables can be set as the wing sweep angle and fuselage height geometric parameters of a hypersonic aircraft.

[0096] Step 2: Determine the baseline aerodynamic layout of the hypersonic vehicle and use the state-type function method to parametrically describe the vehicle's geometric configuration;

[0097] The baseline aerodynamic layout of hypersonic aircraft is based on the SR-72, employing a large-sweep double-delta wing and tailless aerodynamic layout. The fuselage section features a sharp side-edge spine configuration, with double-sweep delta wings on both sides, a pair of elevators on each side controlling the pitch channels, and a single vertical tail controlling the lateral flight path.

[0098] Step 3: Based on the above parametric geometry, high-fidelity and low-fidelity datasets are obtained using engineering estimation and CFD fluid dynamics methods, respectively. Multi-fidelity data fusion is then performed on the high-fidelity and low-fidelity datasets using the Bayesian extended intrinsic orthogonal decomposition method to obtain a surrogate model for predicting the aerodynamic characteristics of the aircraft, such as... Figure 2 As shown;

[0099] Step 31: Initialize the design variables of the multi-confidence aerodynamic model, design experiments for different confidence models, obtain the corresponding model responses, and obtain high and low confidence datasets;

[0100] Assume the design variables are x = [x1, x2, ..., x...]. d ]∈D, where D is the design space, d is the dimension of the design space, and satisfies R is the set of real numbers;

[0101] Within the design space D, different confidence models are sampled to obtain high and low confidence data sets, where the low confidence data set Y:=[y 1 ,...,y n ]=[y(ξ1),...,y(ξ n Design variable ξ for low-reliability data i∈D, i=1,...,n, where n is the number of low-confidence datasets; the high-confidence dataset t:=[t1,...,t s ]=[y′(e1),...,y′(e s High-confidence data design variable e i ∈D, i=1,...,s; s is the number of high-confidence datasets. The sampling method is one of the following: optimal Latin hypercube sampling, full factorial design, or orthogonal experimental design, or it is assumed that the sample points are given.

[0102] Step 32: Based on the low-fidelity data model, calculate the intrinsic orthogonal decomposition mode matrix through singular value decomposition; for the above low-fidelity dataset Y∈R N×n Singular value decomposition yields:

[0103] Y=U∑V T

[0104] In the formula: U=[u 1 ,...,u N ]∈R N×N and V = [v 1 ,...,v n ]∈R n×n It is an orthogonal matrix, i.e., U T U=UU T =I N and V T V = VV T =I n , where ∑=diag(σ1,...,σ n )∈R N×n Includes singular values ​​σ1≥...≥σ in descending order n ≥0;

[0105] Assuming the rank of matrix Y is r = rank(Y), then only the first r singular values ​​are non-zero. The corresponding r left singular vectors, i.e., the first r columns of matrix U, constitute a set of singular values ​​derived from the dataset y1,...,y n The orthonormal basis {u1,...,u} of the space formed r}, that is, the intrinsic orthogonal decomposition basis U r .

[0106] Step 33: Search and determine the coordinates of the low-fidelity data point closest to the high-confidence data. Based on the least squares method and combined with the above-mentioned intrinsic orthogonal decomposition mode matrix, the intrinsic orthogonal decomposition basis coefficients can be obtained.

[0107] A given vector t∈R of high-confidence data s It can be defined as y′∈R N There are only components in the middle. Given a vector, where s < N is the number of sampling points, j1,...,j s ∈{1,...,N}:

[0108]

[0109] For a matrix in Indicates the j-th i There are s standard basis vectors. The components of vector t are represented by the components of vector y′ by a nearest neighbor search of the coordinates of s sampling points in the computational grid;

[0110] The vector y′ is obtained by approximation in the eigenorthogonal decomposition subspace, and the eigenorthogonal decomposition basis coefficients can be found. Make

[0111]

[0112] In the formula, U r =[u 1 ,...,u r ]∈R N×r This is the eigenorthogonal decomposition basis vector matrix, i.e., the first r columns of U. The basis coefficient vectors... The L2 error that minimizes the observations for vector y′ is defined by the least squares problem.

[0113]

[0114] Typically, X = P T U r ∈R s×r Having full column rank, the above equation has a unique solution:

[0115]

[0116] This basic coefficient vector Substituting into the approximate equation, we can obtain an estimate of the vector y′.

[0117] Step 34: Based on Bayesian theory, the least squares problem of eigenorthogonal decomposition is transformed into a regression problem. For the eigenorthogonal decomposition mode matrix and the corresponding high-fidelity data set, a corresponding Gaussian process regression model is constructed.

[0118] Based on the Bayesian extension concept, the least squares problem equation of eigenorthogonal decomposition is defined as a regression process, and a mapping f is constructed: Directly use matrix U r The row is mapped to the vector y′

[0119]

[0120] A set of data pairs is obtained at a given sampling location:

[0121] {(x i ,t i |i=1,...,s}

[0122] in U represents the mode matrix of the eigenorthogonal decomposition. r The jth i OK, For the corresponding sampled response;

[0123] The input-output relationship of the above data set is then surrogate using Gaussian process regression:

[0124] Suppose f(x) is a Gaussian process with zero mean, for any two inputs x, x′∈R r Its covariance Cov[f(x),f(x′)] is given by the covariance function k(x,x′), which adopts a widely used class of covariance functions as shown below:

[0125] k(x,x′)=θ0·exp(-θ1||xx′|| 2 )+θ2x T x′

[0126] In the formula, θ0, θ1, and θ2 are hyperparameters, which are usually not defined in advance but determined from the data by maximizing the log-boundary likelihood function.

[0127]

[0128] So, for a new input variable x * ∈R r f * := f(x * The predicted distribution of ) is

[0129] E[f * ]=k(x * (K+σ) 2 I) -1 t

[0130] Var[f * ]=k(x * ,x * )-k(x * ) T (K+σ 2 I) -1 k(x * )

[0131] Where k(x) * :=(k(x *,x i )) i＝1,..,s ∈R s and K:=(k(x i ,x j )) i,j＝1,..,s ∈R s×s .

[0132] Step 35: By evaluating the predicted distribution of all rows of the intrinsic orthogonal decomposition mode matrix, output the mean and variance information of the data fusion solution under different input conditions;

[0133] By evaluating all rows x of the modality matrix * =(U r ) i The predicted distribution of i = 1, ..., N is used to obtain the data fusion prediction results based on the Bayesian extended intrinsic orthogonal decomposition method.

[0134] The fusion evaluation criteria include the root mean square error (RMSE) and the coefficient of determination (R²). 2 And the interval evaluation criterion PIC:

[0135]

[0136]

[0137]

[0138] Where, N V To verify the sample size, y i To verify the true values ​​of the sample points, To verify the predicted values ​​of the sample points, For y i The average value of q is any quantile between (0,1), typically q = 0.025 and q = 0.975.

[0139] Step 4: Establish a dynamic characteristic model of the aerospace vehicle, and use the aerospace vehicle aerodynamic stability and control co-design method based on the spatial criterion diagram, combined with the flight envelope, to determine whether the vehicle meets the lateral aerodynamic stability requirements under multiple state points.

[0140] The attitude dynamics model of the hypersonic vehicle is constructed as follows:

[0141]

[0142]

[0143] In the formula, (α,β,μ) represent the angle of attack, sideslip angle, and roll angle of the aircraft; (p,q,r) represent the roll rate, pitch rate, and yaw rate; L and Y represent the lift and side force acting on the aircraft, respectively; L, M, and N represent the roll moment, pitch moment, and yaw moment acting on the aircraft, respectively; m and V represent the mass and velocity of the aircraft, respectively; I x I z I y These are the moments of inertia of each axis; I zx For the rotational product of each axis.

[0144] State points are selected based on the flight envelope, and the model is linearized at each state point. Based on the principle of small disturbances, the attitude dynamics equations can be decoupled into longitudinal and lateral motions after linearization using the horizontal no-sideslip flight condition. Ignoring gravity, lift, and lateral forces, the above equations are decoupled into longitudinal and lateral equations. Stability analysis and control system design are based on the following equations:

[0145]

[0146]

[0147] In the formula, L′ α L′ β L′ p L r ′、L q ′、 These are the derivatives of the rolling torque for each channel; N′ α 、N′ β 、N′ p N r ′、N q ′、 These are the derivatives of the yaw moment for each channel; M′ α M′ β , M q ′ represents the derivative of the pitch moment for each channel; δ e Elevator deflection angle; δ a This refers to the aileron deflection angle.

[0148] The operators in the equation are defined as follows:

[0149]

[0150] For hypersonic vehicles, the lateral path is usually unstable, and the selection of lateral stability parameters and lateral control laws is particularly important.

[0151] By solving the characteristic equation |λI-A|, the lateral stability criterion can be written as the following equation:

[0152]

[0153] In the formula, C nβ C lβ These are two classic stability parameters.

[0154] Based on the spatial criterion graph method, the x-axis and y-axis are respectively mapped to the lateral stability parameter C. lβ and C nβ The above equation can be written as:

[0155]

[0156] This allows us to draw a spatial criterion diagram for lateral open-loop stability, determining whether the lateral stability requirements are met at multiple state points within the flight envelope of the aircraft under this configuration. If not, we select new geometric parameters and return to step 2; if they are met, we proceed to step 5.

[0157] Step 5: Under the premise of meeting the above stability requirements, introduce a control system and combine it with control law design. By adjusting the controller parameters, meet the closed-loop control aerodynamic stability criteria and non-minimum phase requirements, and finally output the aerodynamic layout design results of the aircraft that meet the requirements of aerodynamic stability and control coordinated design.

[0158] Since the lateral path of the aircraft is unstable, a control law is required for stabilization. Therefore, the design of the control system is introduced, and the closed-loop control aerodynamic stability criterion is as follows:

[0159]

[0160] In the formula, H and F represent different closed-loop stability regions.

[0161] Similarly, this corresponds to the spatial criterion diagram:

[0162]

[0163] Furthermore, according to the transverse non-minimum phase criterion, there exists a straight line passing through the origin:

[0164]

[0165] In the formula, These are the aerodynamic stability parameters caused by the ailerons.

[0166] When the stability parameter (C) lβ C nβWhen the point corresponding to the given condition is located above this straight line, it indicates that the lateral system of the aircraft is a non-minimum phase system and can be controlled using conventional control methods. In summary, the final output satisfies the aerodynamic layout design results of the aircraft under the requirements of aerodynamic stability and control co-design.

Claims

1. A collaborative design method for the overall control of a hypersonic vehicle based on multi-fidelity data fusion, characterized in that, Includes the following steps: Step 1: Initialize the aircraft's geometric design variables and constraints; Step 2: Determine the baseline aerodynamic layout of the hypersonic vehicle and use the state-type function method to parametrically describe the vehicle's geometric configuration; Step 3: Based on the above parametric geometry, high-fidelity and low-fidelity datasets are obtained using engineering estimation and CFD fluid dynamics methods, respectively. Multi-fidelity data fusion is then performed on the high-fidelity and low-fidelity datasets using the Bayesian extended intrinsic orthogonal decomposition method to obtain a surrogate model for predicting aircraft aerodynamic characteristics. The Bayesian extended intrinsic orthogonal decomposition data fusion method specifically includes the following steps: Step 31: Initialize the design variables of the multi-confidence aerodynamic model, design experiments for different confidence models, obtain the corresponding model responses, and obtain high and low confidence datasets; Step 32: Based on the low-confidence data model, the intrinsic orthogonal decomposition mode matrix is ​​obtained by singular value decomposition. Step 33: Search and determine the coordinates of the low-fidelity data point closest to the high-confidence data, and obtain the intrinsic orthogonal decomposition basis coefficients based on the least squares method and the aforementioned intrinsic orthogonal decomposition mode matrix; Step 34: Based on Bayesian theory, the least squares problem of eigenorthogonal decomposition is transformed into a regression problem. For the eigenorthogonal decomposition mode matrix and the corresponding high-fidelity data set, a corresponding Gaussian process regression model is constructed. Step 35: By evaluating the predicted distribution of all rows of the intrinsic orthogonal decomposition mode matrix, output the mean and variance information of the data fusion solution under different input conditions; Step 4: Establish a dynamic characteristic model of the aerospace vehicle, and use the aerospace vehicle aerodynamic stability and control co-design method based on the spatial criterion diagram, combined with the flight envelope, to determine whether the vehicle meets the lateral aerodynamic stability requirements under multiple state points. Step 5: Under the premise of meeting the above stability requirements, introduce a control system and combine it with control law design. By adjusting the controller parameters, meet the closed-loop control aerodynamic stability criteria and non-minimum phase requirements, and finally output the aerodynamic layout design results of the aircraft that meet the requirements of aerodynamic stability and control co-design.

2. The hypersonic vehicle overall control collaborative design method based on multi-fidelity data fusion as described in claim 1, characterized in that, In step 2, the hypersonic vehicle adopts a hypersonic aircraft with a large swept double delta wing and no horizontal tail aerodynamic layout. The fuselage cross section has a sharp side edge spine configuration, and the two sides of the fuselage are double swept delta wings. A pair of elevators on both sides control the pitch channel, and a single vertical tail design controls the lateral channel of flight.

3. The hypersonic vehicle overall control collaborative design method based on multi-fidelity data fusion as described in claim 1, characterized in that, In step 31, the design variables of the multi-confidence aerodynamic model are initialized, and experiments are designed for different confidence levels to obtain the corresponding model responses, thus obtaining the high and low confidence datasets. This process specifically includes the following steps: Step 31a, design variables are ,in, For design space, To design the spatial dimension, and satisfy , It is the set of real numbers; Step 32b, Experimental design refers to designing a space... The system samples data from different credibility models to obtain high and low credibility datasets, with the low credibility dataset being... Design variables for low-reliability data , The number of low-confidence datasets; the number of high-confidence datasets. High-reliability data design variables ; The number of high-confidence datasets is denoted by the sampling method, which is one of the following: optimal Latin hypercube sampling, full factorial design, or orthogonal experimental design, or it is assumed that the sample points are given.

4. The hypersonic vehicle overall control collaborative design method based on multi-fidelity data fusion as described in claim 1, characterized in that, In step 32, based on the low-confidence data model, the intrinsic orthogonal decomposition mode matrix is ​​calculated through singular value decomposition as follows: For the aforementioned low-fidelity dataset Singular value decomposition yields: In the formula, and It is an orthogonal matrix, that is and ,in Contains singular values ​​in descending order ; Assumption Matrix The rank is Then only the first Each singular value is non-zero, and the corresponding... Left singular vectors, i.e., matrices The former Columns constitute a dataset The standard orthogonal basis of the space formed That is, the intrinsic orthogonal decomposition basis .

5. The hypersonic vehicle overall control collaborative design method based on multi-fidelity data fusion as described in claim 1, characterized in that, In step 33, the coordinates of the low-fidelity data points closest to the high-confidence data are determined. Based on the least squares method and combined with the aforementioned intrinsic orthogonal decomposition mode matrix, the intrinsic orthogonal decomposition basis coefficients can be obtained as follows: A given vector of high-confidence data It can be defined as being in There are only components in the middle. Given vectors, where, The number of sampling points. : For a matrix ,in , Indicates the first A standard basis vector, obtained by applying it in the computational grid. Nearest neighbor search for the coordinates of each sampling point, using vectors The components are used to characterize the vector The amount; vector The eigenorthogonal decomposition basis coefficients can be found by approximation in the eigenorthogonal subspace. , making In the formula, The eigenorthogonal decomposition basis vector matrix is, i.e. The former Columns, basis coefficient vector Generate for vector The smallest observation item Error, defined by the least squares problem generally, Having full column rank, the above equation has a unique solution: This basic coefficient vector Substituting into the approximate equation, we obtain the vector. The estimated value .

6. The hypersonic vehicle overall control collaborative design method based on multi-fidelity data fusion as described in claim 1, characterized in that, In step 34, based on Bayesian theory, the least squares problem of eigenorthogonal decomposition is transformed into a regression problem. Specifically, for the eigenorthogonal decomposition mode matrix and the corresponding high-fidelity dataset, a corresponding Gaussian process regression model is constructed as follows: Based on the Bayesian extension concept, the least squares problem equation of eigenorthogonal decomposition is defined as a regression process, and a mapping is constructed. Directly put the matrix Rows mapped to vectors A set of data pairs is obtained at a given sampling location: in The mode matrix representing the eigenorthogonal decomposition The OK, The corresponding sampled response; The input-output relationship of the above data set is then surrogate using Gaussian process regression: Assumption A Gaussian process with zero mean, for any two inputs Its covariance From covariance function The following is an example of a widely used covariance function: In the formula, , , Hyperparameters are typically not defined in advance, but rather determined from the data by maximizing the log-boundary likelihood function. So, for a new input variable , The predicted distribution is in and .

7. The hypersonic vehicle overall control collaborative design method based on multi-fidelity data fusion as described in claim 1, characterized in that, In step 35, by evaluating the predicted distribution of all rows of the intrinsic orthogonal decomposition mode matrix, the mean and variance information of the data fusion solution under different input conditions are output. Specifically, this involves evaluating all rows of the mode matrix. The predicted distribution is obtained, and the data fusion prediction results based on the Bayesian extended eigenorthogonal decomposition method are obtained. ; The fusion evaluation criteria include the root mean square error (RMSE) and the coefficient of determination (R²). 2 And the interval evaluation criterion PIC: in, To verify the sample size, To verify the true values ​​of the sample points, To verify the predicted values ​​of the sample points, for The average value, for Take any quantile between 1 and 2. and .

8. The hypersonic vehicle overall control collaborative design method based on multi-fidelity data fusion as described in claim 1, characterized in that, In step 4, a dynamic characteristic model of the aerospace vehicle is established. Based on the space criterion diagram, a collaborative design method for aerospace vehicle aerodynamic stability and control is used. Combined with the flight envelope, it is determined whether the vehicle meets the lateral aerodynamic stability requirements at multiple state points. Specifically: The attitude dynamics model of the hypersonic vehicle is constructed as follows: In the formula, For the aircraft's angle of attack, sideslip angle, and roll angle; For roll rate, pitch rate, and yaw rate, , These are the lift and lateral forces acting on the aircraft, respectively. , , These are the rolling moment, pitching moment, and yaw moment acting on the aircraft, respectively. , These refer to the aircraft's mass and speed, respectively. , , These are the moments of inertia of each axis; For the rotational product of each axis; State points are selected based on the flight envelope, and the model is linearized at each state point; Based on the principle of small disturbances, the attitude dynamics equations are linearized and decoupled into longitudinal motion and lateral motion by utilizing the horizontal no-sideslip flight condition; Neglecting gravity, lift, and lateral forces, the above equations are decoupled into longitudinal and lateral equations. Stability analysis and control system design are based on the following equations: In the formula, , , , , , These are the derivatives of the rolling torque for each channel; , , , , , These are the derivatives of the yaw moment for each channel; , , , These are the derivatives of the pitch moment for each channel; This refers to the elevator deflection angle; For aileron deflection angle; The operators in the equation are defined as follows: For hypersonic vehicles, the lateral channel is unstable, and the selection of lateral stability parameters and lateral control laws is particularly important. By solving the characteristic equation The lateral stability criterion is written as the following equation: In the formula, , These are two classic stability parameters; Based on the spatial criterion graph method, the x-axis and y-axis are respectively mapped to the lateral stability parameters. and The above equation can be written as: This allows us to draw a spatial criterion diagram for lateral open-loop stability, determining whether the lateral stability requirements are met at multiple state points within the flight envelope of the aircraft under this configuration. If not, we select new geometric parameters and return to step 2; if they are met, we proceed to step 5.

9. The hypersonic vehicle overall control collaborative design method based on multi-fidelity data fusion as described in claim 1, characterized in that, In step 5, under the premise of satisfying the above stability requirements, a control system is introduced in conjunction with the control law design. By adjusting the controller parameters, the closed-loop control aerodynamic stability criterion and non-minimum phase requirements are met. The final output is the aerodynamic layout design result of the aircraft that meets the requirements of aerodynamic stability and control coordinated design. Since the lateral path of the aircraft is unstable, a control law is required for stabilization. Therefore, the design of the control system is introduced, and the closed-loop control aerodynamic stability criterion is as follows: In the formula, H and F represent different closed-loop stability regions; Similarly, this corresponds to the spatial criterion diagram: Furthermore, according to the transverse non-minimum phase criterion, there exists a straight line passing through the origin: In the formula, , These are the aerodynamic stability parameters caused by the ailerons; When stability parameter When the corresponding point is located above this straight line, it indicates that the lateral system of the aircraft is a non-minimum phase system, and it can be controlled by conventional control methods.

Images