Aircraft layout generative design method based on multi-model cooperation

Abstract

This invention discloses a generative design method for aircraft layout based on multi-model collaboration. The method first transforms unstructured requirements into structured objectives using a requirement analysis model and extracts feasible domain boundaries. Second, based on differentiated preferences, it uses parallel instantiation to generate parameters using a parameter generation model, and after over-boundary projection correction, obtains multi-scale initial parameters including micro-airfoil and macro-layout parameters. Next, it performs aerodynamic simulation and geometric verification to obtain performance data. Then, it constructs a cross-scale mapping to calculate the sensitivity matrix, combines performance deviation and constraint surrogate models to determine the optimization direction, and calculates adaptive damping coefficients to generate corrected gradients. Finally, it performs cross-scale joint step size correction and closed-loop iteration until convergence to obtain candidate parameters, and comprehensively decides and outputs the optimal solution. This invention achieves accurate multi-scale parameter correction driven by performance deviation, improving convergence and practicality.

Description

Technical Field

[0001] This invention relates to the field of aircraft aerodynamic layout design technology, and in particular to a generative design method for aircraft layout based on multi-model collaboration. Background Technology

[0002] Aerodynamic layout design is a core aspect of overall aircraft design, directly determining its aerodynamic performance and engineering practicality. Aerodynamic layout is composed of both microscopic airfoil parameters and macroscopic layout parameters. Macroscopic layout parameters dominate the overall spatial flow field distribution and global aerodynamic load characteristics, while microscopic airfoil parameters control local boundary layer flow and pressure gradient distribution. These two aspects exhibit significant differences in magnitude and strong coupling effects in their aerodynamic responses. Traditional aircraft layout optimization design heavily relies on the engineering experience of senior designers for multi-scale parameter decoupling, employing a sequential design and trial-and-error correction approach. This approach suffers from inherent drawbacks such as the difficulty of multidisciplinary coupled optimization, long design cycles, and difficulty in guaranteeing global optimality.

[0003] With the development of generative artificial intelligence and automated optimization technology, model-based parametric generation and closed-loop iterative design have provided a new paradigm for aircraft layout design, effectively improving the efficiency of design space exploration. However, existing technologies still have significant shortcomings in handling the collaborative iterative optimization of micro and macro multi-scale parameters. They lack a physical mechanism to accurately decouple the overall aerodynamic performance deviations and explicitly map them to design parameters at different scales. In the optimization process driven by performance deviations, macro parameters are usually much more sensitive to aerodynamic indicators than micro parameters, which can easily lead to gradient updates being dominated by macro components. This can cause a scale interference phenomenon where macro parameters rapidly converge to the boundary while micro parameter updates stagnate. At the same time, the parameter correction process lacks cross-scale collaborative constraints based on the continuity of flow field evolution and geometric deformation. Independent adjustments of micro and macro parameters can easily disrupt the overall physical coupling relationship of the aerodynamic layout, leading to local flow separation or geometric distortion. This lack of deviation tracing and scale response imbalance makes the optimization search direction blind, severely restricting the convergence efficiency of closed-loop iterative optimization.

[0004] Therefore, there is an urgent need to construct a generative design method for aircraft layout with cross-scale physical mapping and adaptive collaborative correction capabilities to solve the core problem of convergence difficulties caused by multi-scale parameter iterative interference and blind optimization in existing technologies. Summary of the Invention

[0005] To address these issues, this invention provides a generative design method for aircraft layout based on multi-model collaboration, thereby resolving the aforementioned problems in the prior art.

[0006] To achieve the above objectives, this invention provides a generative design method for aircraft layout based on multi-model collaboration, comprising:

[0007] Step S1: Based on the requirement analysis model, the unstructured requirements of the aircraft layout are transformed to obtain structured design objectives, and the boundaries of the structured design objectives are extracted to obtain the parameter feasible domain boundaries.

[0008] Step S2: Based on the differentiated preference configuration, the structured design objective is weighted to instantiate multiple design processes in parallel; the parameter generation model is called to generate design parameters to be verified based on the structured design objective and the differentiated preference configuration; the design parameters to be verified are boundary-checked and over-boundary projection corrected based on the feasible domain boundary to obtain initial design parameters that satisfy the feasible domain, wherein the initial design parameters include multi-scale design parameters composed of micro airfoil parameters and macro layout parameters;

[0009] Step S3: Perform aerodynamic performance simulation and geometric constraint verification based on the initial design parameters to obtain performance indicators and simulation characteristic data;

[0010] Step S4: Based on the simulation feature data and the multi-scale design parameters, construct a cross-scale mapping relationship to calculate the sensitivity of the performance index to the design parameters and obtain a sensitivity matrix; calculate the performance deviation based on the performance index and the structured design objective; based on the preset constraint surrogate model, use the performance deviation as the objective function and combine it with the sensitivity matrix to determine the optimization direction, and calculate the adaptive damping coefficient of each scale parameter to generate the parameter correction gradient.

[0011] Step S5: Based on the parameter correction gradient and the adaptive damping coefficient, perform cross-scale joint step size correction on the micro airfoil parameters and the macro layout parameters to obtain updated design parameters; repeat steps S3 to S5 based on the updated design parameters until the convergence criterion is met to obtain candidate design parameters.

[0012] Step S6: Based on the candidate design parameters of each group and their corresponding differentiated preference configurations, a comprehensive decision is made and the best option is selected to obtain the optimal aircraft layout design scheme.

[0013] Furthermore, the process of extracting the boundaries of the structured design objective to obtain the boundary of the parameter feasible region includes:

[0014] Based on the preset engineering constraint specifications, the boundary conditions of the structured design objective are solved, and the engineering constraint specifications are mapped to the parameter interval limits corresponding to the structured design objective to construct the parameter feasible domain boundary.

[0015] Furthermore, the process of step S2 includes:

[0016] Step S21: Based on the preset differential preference configuration, determine the preference weight vector for lift coefficient, lift-to-drag ratio and pitch static stability derivative, and instantiate multiple independent design processes in parallel according to the preference weight vector;

[0017] Step S22: Use the structured design objective and the preference weight vector as the conditional input of the parameter generation model to generate design parameters to be verified, including airfoil feature vector and wing layout vector;

[0018] Step S23: Match the design parameters to be verified with the boundary of the feasible region of parameters. If there are out-of-bounds parameters that exceed the boundary of the feasible region of parameters, then project and truncate or correct the out-of-bounds parameters along the normal of the feasible region constraint hyperplane, so that all design parameters to be verified converge within the feasible region of parameters, and obtain the initial design parameters.

[0019] Furthermore, the process of step S23 includes:

[0020] The design parameters to be verified are subjected to dimension-by-dimensional comparison and cross-scale coupling verification.

[0021] If a single-dimensional parameter exceeds its independent boundary, it is truncated by projection along the normal of the feasible region constraint hyperplane, and the single-dimensional parameter is forced to the boundary extremum of its out-of-bounds direction, while the other dimensional parameters remain unchanged.

[0022] If the multidimensional parameters violate the cross-scale coupling constraint, a neighborhood mapping correction is performed along the normal of the feasible region constraint hyperplane. An affine mapping matrix is ​​constructed based on the normal gradient of the hyperplane at the current boundary point, and the multidimensional parameters are collaboratively scaled or translated into the feasible region along the physical association direction determined by the affine mapping matrix.

[0023] Furthermore, the process of step S4 includes:

[0024] Step S41: Based on the simulation feature data, solve the partial derivatives of the performance index with respect to the multi-scale design parameters to construct a cross-scale sensitivity matrix;

[0025] Step S42: Calculate the performance deviation between the performance index and the structured design objective;

[0026] Step S43: Apply a penalty term to the performance deviation based on the physical constraint proxy model, and determine the constrained optimization direction by combining the cross-scale sensitivity matrix;

[0027] Step S44: Calculate the adaptive damping coefficient based on the magnitude difference of the partial derivatives of the parameters at each scale in the cross-scale sensitivity matrix, and generate the parameter correction gradient in combination with the constrained optimization direction.

[0028] Furthermore, the process of step S44 includes:

[0029] The first partial derivative norm of the corresponding micro airfoil parameter and the second partial derivative norm of the corresponding macro layout parameter are extracted from the cross-scale sensitivity matrix, respectively.

[0030] Based on the ratio of the first partial derivative norm to the second partial derivative norm, a nonlinear decay function is used to calculate the adaptive damping coefficient ratio between the micro airfoil parameters and the macro layout parameters, so that parameters with smaller partial derivatives can obtain larger damping coefficients.

[0031] The constrained optimization direction is scaled across scales according to the adaptive damping coefficient ratio to generate a parameter correction gradient with differentiated step sizes.

[0032] Furthermore, the process of step S5 includes:

[0033] Step S51: Decompose the parameter correction gradient into micro airfoil gradient and macro layout gradient, and scale them according to their respective adaptive damping coefficients to obtain micro airfoil correction step size and macro layout correction step size.

[0034] Step S52: Based on the correction step size, the micro airfoil parameters and macro layout parameters are updated synchronously; if the updated parameters exceed the boundary of the parameter feasible region, the micro airfoil correction step size and the macro layout correction step size are reduced in the same proportion until the updated parameters fall into the feasible region to obtain the updated design parameters.

[0035] Step S53: When the aerodynamic performance deviation of the micro airfoil profile and the aerodynamic performance deviation of the macro layout of the whole aircraft simultaneously meet their respective preset tolerances, convergence is determined; if the performance deviation does not decrease in a set number of consecutive iterations, an early stop mechanism is triggered, and the historical optimal parameters are output as the candidate design parameters.

[0036] Further, the process of step S52 includes:

[0037] If the macro-layout parameters conform to the feasible domain boundary of the parameters after the collaborative reduction, and the current performance deviation is still greater than the preset convergence threshold, then it is determined that the macro-layout optimization space is exhausted.

[0038] Freeze the macroscopic layout parameters, remove the same proportional collaborative reduction constraint, and perform independent iterative updates on the micro airfoil parameters based only on the micro airfoil correction step size until the updated parameters fall within the feasible region.

[0039] Furthermore, step S6 includes the following process:

[0040] Step S61: Obtain the aerodynamic performance deviation and geometric constraint violation degree corresponding to each group of candidate design parameters;

[0041] Step S62: Calculate the hard constraint penalty factor based on the geometric constraint violation degree;

[0042] Step S63: Based on the weight vector corresponding to the differentiated preference configuration, nonlinear weighting is applied to the aerodynamic performance deviation to calculate the preference fit.

[0043] Step S64: Combine the preference fit with the hard constraint penalty factor to calculate the overall fitness.

[0044] Step S65: Select the candidate design parameters with the best overall adaptability as the optimal aircraft layout design scheme.

[0045] Compared with existing technologies, the advantages of this invention lie in its ability to accurately map aerodynamic performance deviations to macroscopic layout and microscopic airfoil parameters by constructing a cross-scale sensitivity matrix, establishing a physical causal relationship between aerodynamic response and geometric deformation, and overcoming the shortcomings of traditional methods in terms of blind optimization direction. For aerodynamic characteristics characterized by high global sensitivity of macroscopic parameters and low local sensitivity of microscopic parameters, an adaptive damping mechanism is introduced to balance the multi-scale correction amplitude, ensuring synchronous coordination between the evolution of the overall flow field and the aerodynamic characteristics of the airfoil profile. During step-size correction, cross-scale collaborative reduction and independent microscopic iteration under macroscopic freezing are implemented based on boundary constraints, maintaining the continuous stability of geometric deformation and physical response during the optimization process and avoiding parameter overshoot and scale interference. The combination of parallel exploration and comprehensive decision-making based on differentiated preferences effectively resolves the limitations of optimization under multi-objective conflicts. This method achieves accurate closed-loop correction of multi-scale geometric parameters driven by aerodynamic performance deviations, significantly improving the global convergence capability and engineering practicality of aircraft layout design. Attached Figure Description

[0046] Figure 1 A flowchart illustrating the generative design method for aircraft layout based on multi-model collaboration provided by this invention;

[0047] Figure 2 This is a flowchart illustrating step S2 in the multi-model collaborative generative design method for aircraft layout provided by the present invention.

[0048] Figure 3 The three-view diagrams of a high-altitude long-endurance flying-wing UAV model generated for an embodiment of the present invention. Detailed Implementation

[0049] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0050] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0051] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0052] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0053] Please see Figure 1 As shown, this invention provides a generative design method for aircraft layout based on multi-model collaboration, including:

[0054] Step S1: Based on the requirement analysis model, the unstructured requirements of the aircraft layout are transformed to obtain structured design objectives, and the boundaries of the structured design objectives are extracted to obtain the parameter feasible domain boundaries.

[0055] Specifically, the structured design objectives include quantitative indicators such as aircraft type, takeoff weight, cruise speed, and lift-to-drag ratio; based on the aircraft type and aerodynamic layout, the parameter dimensions of the boundary to be extracted are determined, including micro airfoil parameters (such as shape weight coefficients of each order in the CST parameterization method, maximum relative thickness) and macro layout parameters (such as aspect ratio, root-to-tip ratio, leading edge sweep angle, and wing reference area). Extract the airfoil geometric effectiveness criteria and aerodynamic sensitivity criteria from the preset engineering constraints: Based on the geometric compliance constraint that "airfoil thickness is positive everywhere and decreases along the spanwise direction," set a lower limit for the maximum relative thickness of the airfoil (e.g., set the lower limit of the maximum relative thickness of the root airfoil to 12% to meet structural volume requirements); based on the sensitivity of the Reynolds number to frictional drag under cruise conditions, set an upper limit for the maximum relative thickness (e.g., set the upper limit of the maximum relative thickness of the wingtip airfoil to 10% to suppress boundary layer thickening); simultaneously, for the weighting coefficients of each order of shape function in the airfoil CST parametric expression, based on the compliance constraint to prevent self-intersection of airfoil surface curvature, monitor the surface curvature by perturbing the extreme values ​​of the weighting coefficients, and calibrate the independent value range of each order of weighting coefficients. Based on the aircraft type statistical specifications in the aviation knowledge base, the initial statistical intervals of macroscopic layout parameters that match the structured design objectives are retrieved. For example, for high-altitude long-endurance flying wing layouts, the aspect ratio interval AR∈[8.0,10.0], the root-to-tip ratio interval λ∈[0.3,0.6], and the leading-edge sweep angle interval Λ∈

[20] are extracted. ∘ 40 ∘ On the other hand, the boundary is solved based on aerodynamic-structural coupling constraints, according to the takeoff weight. With cruising speed Combining the maximum lift coefficient limit and stall speed safety specifications, the lift balance equations are used. Computer Wing Reference Area Feasible lower bound Where ρ is the air density, For cruising speed, This is the maximum lift coefficient.

[0056] Specifically, the process of extracting the boundaries of the structured design objective to obtain the boundary of the parameter feasible domain includes:

[0057] Based on the preset engineering constraint specifications, the boundary conditions of the structured design objective are solved, and the engineering constraint specifications are mapped to the parameter interval limits corresponding to the structured design objective to construct the parameter feasible domain boundary.

[0058] The independent value ranges of the micro airfoil parameters are combined with the statistical ranges of the macro layout parameters and the lower bound of the equation solution, and cross-scale geometric correlation constraints (such as aspect ratio definition constraints) are introduced. (where b is the wingspan). By combining the above independent interval boundaries with the associated constraint equations, a multidimensional parametric feasible region boundary Ω expressed by a system of analytical inequalities is constructed:

[0059]

[0060] Where x is a multidimensional design variable vector containing micro airfoil parameters and macro layout parameters. and These are the upper and lower limits of an independent interval. This refers to the constraints of cross-scale coupling and physical equilibrium.

[0061] Step S2: Based on the differentiated preference configuration, the structured design objective is weighted to instantiate multiple design processes in parallel; the parameter generation model is called to generate design parameters to be verified based on the structured design objective and the differentiated preference configuration; the design parameters to be verified are boundary-checked and over-boundary projection corrected based on the feasible domain boundary to obtain initial design parameters that satisfy the feasible domain, wherein the initial design parameters include multi-scale design parameters composed of micro airfoil parameters and macro layout parameters;

[0062] Specifically, such as Figure 2 As shown, the process of step S2 includes:

[0063] Step S21: Based on the preset differential preference configuration, determine the preference weight vector for lift coefficient, lift-to-drag ratio and pitch static stability derivative, and instantiate multiple independent design processes in parallel according to the preference weight vector;

[0064] Specifically, differentiated preference configurations are extracted based on the aircraft mission scenario, and these preference configurations are quantized into a normalized preference weight vector. ,in For example, configuring preference weight vectors for high-altitude long-endurance missions. =[0.2,0.7,0.1] This is designed to guide a high lift-to-drag ratio, with a preferred weight vector configured for high-maneuverability missions. =[0.6,0.2,0.2] to guide high-lift design; in the computational framework, an independent computation thread or GPU core is assigned to each set of preference weight vectors to build multiple parallel and non-interfering design processes.

[0065] Step S22: Use the structured design objective and the preference weight vector as the conditional input of the parameter generation model to generate design parameters to be verified, including airfoil feature vector and wing layout vector;

[0066] Specifically, the structured design objectives (such as target cruise Mach number, design lift coefficient) are concatenated with the preference weight vector w of each process group to form a conditional vector c = [target index, w]. This conditional vector c, along with random latent variables z sampled from a standard normal distribution, is input into the decoder of a pre-trained conditional variational autoencoder (CVAE) or conditional diffusion model. The decoder outputs a multidimensional continuous vector, which is then decomposed into micro-airfoil parameters (such as the weight coefficients of each order in the CST parameterization method). Maximum relative thickness The design parameters to be verified are formed by combining macroscopic layout parameters (such as aspect ratio AR, root-to-tip ratio λ, and leading-edge sweep angle Λ). .

[0067] Step S23: Match the design parameters to be verified with the boundary of the feasible region of parameters. If there are out-of-bounds parameters that exceed the boundary of the feasible region of parameters, then project and truncate or correct the out-of-bounds parameters along the normal of the feasible region constraint hyperplane, so that all design parameters to be verified converge within the feasible region of parameters, and obtain the initial design parameters.

[0068] Specifically, step S23 includes the following processes:

[0069] The design parameters to be verified are subjected to dimension-by-dimensional comparison and cross-scale coupling verification.

[0070] If a single-dimensional parameter exceeds its independent boundary, it is truncated by projection along the normal of the feasible region constraint hyperplane, and the single-dimensional parameter is forced to the boundary extremum of its out-of-bounds direction, while the other dimensional parameters remain unchanged.

[0071] Specifically, if a single-dimensional parameter Beyond its independent boundaries Then, the projection is truncated along the normal of the independent boundary constraint hyperplane (i.e., the direction of the coordinate axis in that dimension), and the single-dimensional parameter is forced to be the boundary extremum of its out-of-bounds direction. or The other dimensional parameters remain unchanged, that is .

[0072] If the multidimensional parameters violate the cross-scale coupling constraint, a neighborhood mapping correction is performed along the normal of the feasible region constraint hyperplane. An affine mapping matrix is ​​constructed based on the normal gradient of the hyperplane at the current boundary point, and the multidimensional parameters are collaboratively scaled or translated into the feasible region along the physical association direction determined by the affine mapping matrix.

[0073] Specifically, if multidimensional parameters violate cross-scale coupling constraints (For example, wingtip stall constraints or divergent velocity constraints caused by the combination of aspect ratio, sweep angle, and airfoil thickness), then perform neighborhood mapping correction: calculate the current out-of-bounds point. The cross-scale coupling constraint hyperplane described above normal gradient Based on the normal gradient, an affine mapping matrix is ​​constructed to determine the physical association direction. The multidimensional parameters are then cooperatively scaled or translated along the inverse direction of the gradient, i.e., the physical association direction. The specific correction formula is as follows: Where α is the step size factor determined through a one-dimensional linear search, such that The parameters may fall within the feasible region, thus ensuring that the aero-geometric coupling relationship between multiple parameters is not disrupted; after the projection truncation or neighborhood mapping correction, the initial design parameters that satisfy the boundary of the feasible region of the parameters are output.

[0074] Step S3: Perform aerodynamic performance simulation and geometric constraint verification based on the initial design parameters to obtain performance indicators and simulation characteristic data;

[0075] Specifically, multi-scale geometric reconstruction and computational domain discretization are performed based on the initial design parameters. Microscopic airfoil parameters (such as CST weighting coefficients and maximum relative thickness) are analytically converted into two-dimensional airfoil coordinate points through parametric equations. Combined with macroscopic layout parameters (aspect ratio, root-to-tip ratio, leading-edge sweep angle), a three-dimensional airfoil surface is generated through spanwise interpolation and affine transformation. A full-aircraft computational mesh is generated based on the three-dimensional airfoil surface, and mesh refinement is performed on the airfoil surface and wake region to capture boundary layer and wake characteristics. The cruise Mach number, Reynolds number, and angle of attack range corresponding to the structured design objectives are set as far-field boundary conditions. A computational fluid dynamics solver (such as a RANS solver based on the Reynolds-averaged Navier-Stokes equations) is invoked to iteratively solve the flow field on the computational mesh until the residuals converge. The pressure and friction distributions on the entire airfoil surface are extracted, and the macroscopic lift coefficient of the entire airfoil is obtained through area integration. drag coefficient (Then calculate the lift-to-drag ratio K= / ), and torque coefficient (and then calculate the pitch static stability derivative) Simultaneously, the cross-sectional pressure distribution of each spanwise section of the wing is extracted as a micro-aerodynamic feature of the airfoil; based on the generated three-dimensional wing surface, the geometric feature parameters of each spanwise section are extracted to verify the compliance constraint that "the airfoil thickness is positive everywhere and monotonically decreases along the spanwise direction"; the total aircraft volume is calculated and the fuel volume and structural layout space constraints are verified; the wingtip torsional deformation constraints and divergence velocity constraints derived from the macro-layout and micro-airfoil thickness are verified; the degree of violation of each constraint is quantified, with absolute satisfaction recorded as 0 and the out-of-bounds amount recorded as a normalized positive value, generating a geometric constraint violation vector; the aerodynamic coefficients are combined into a performance index vector. The geometric constraint violation vector, key features of the airfoil surface pressure distribution (such as shock wave location coordinates and suction peak size), and flow field characteristic parameters are combined into simulation feature data. The simulation feature data is used for gradient backpropagation and correction optimization of micro airfoils and macro layout in subsequent steps.

[0076] Step S4: Based on the simulation feature data and the multi-scale design parameters, construct a cross-scale mapping relationship to calculate the sensitivity of the performance index to the design parameters and obtain a sensitivity matrix; calculate the performance deviation based on the performance index and the structured design objective; based on the preset constraint surrogate model, use the performance deviation as the objective function and combine it with the sensitivity matrix to determine the optimization direction, and calculate the adaptive damping coefficient of each scale parameter to generate the parameter correction gradient.

[0077] Specifically, step S4 includes the following process:

[0078] Step S41: Based on the simulation feature data, solve the partial derivatives of the performance index with respect to the multi-scale design parameters to construct a cross-scale sensitivity matrix;

[0079] Specifically, based on the simulation feature data, the partial derivatives of the performance indicators with respect to multi-scale design parameters are solved to construct a cross-scale sensitivity matrix: based on the simulation feature data (surface pressure distribution, friction distribution, and flow field characteristics) obtained in step S3, the aerodynamic performance indicators (lift coefficient) are calculated using the discrete adjoint method or differential differentiation based on the chain rule. Lift-to-drag ratio K, pitch static stability derivative Design parameters to be verified The partial derivatives are calculated as follows: for micro airfoil parameters (such as CST weighting coefficients), the derivative of the airfoil normal disturbance is solved by integrating the derivative of the local surface pressure; for macroscopic layout parameters (such as aspect ratio and sweep angle), the derivative of the macroscopic geometric mapping variables is solved by integrating the global aerodynamic forces; all partial derivatives are combined to construct a cross-scale sensitivity matrix with a dimension equal to the number of performance indicators multiplied by the dimension of the design parameters. .

[0080] Step S42: Calculate the performance deviation between the performance index and the structured design objective;

[0081] Specifically, obtain the current performance metric vector J output in step S3. Combined with the preference weight vector determined in step S21 Calculate its relationship with the structured design objective vector. Weighted performance deviation between ;

[0082] Step S43: Apply a penalty term to the performance deviation based on the physical constraint proxy model, and determine the constrained optimization direction by combining the cross-scale sensitivity matrix;

[0083] Specifically, a pre-built physical constraint proxy model is invoked, which outputs the geometric constraint violation vector from step S33. For input, output a scalar penalty term. , where μ is the dynamic penalty factor; construct a comprehensive objective function that includes performance bias and penalty term. Combined with the cross-scale sensitivity matrix G, calculate the unconstrained natural gradient. Determine the constrained optimization direction .

[0084] Step S44: Calculate the adaptive damping coefficient based on the magnitude difference of the partial derivatives of the parameters at each scale in the cross-scale sensitivity matrix, and generate the parameter correction gradient in combination with the constrained optimization direction.

[0085] Specifically, step S44 includes the following process:

[0086] The first partial derivative norm of the corresponding micro airfoil parameter and the second partial derivative norm of the corresponding macro layout parameter are extracted from the cross-scale sensitivity matrix, respectively.

[0087] Specifically, the cross-scale sensitivity matrix G or natural gradient is extracted. The L2 norm of the subvector corresponding to the micro airfoil parameters is calculated as the first partial derivative norm. Extract the sub-vectors corresponding to the macroscopic layout parameters and calculate their L2 norm as the second partial derivative norm. .

[0088] Based on the ratio of the first partial derivative norm to the second partial derivative norm, a nonlinear decay function is used to calculate the adaptive damping coefficient ratio between the micro airfoil parameters and the macro layout parameters, so that parameters with smaller partial derivatives can obtain larger damping coefficients.

[0089] Specifically, based on the ratio of the first partial derivative norm to the second partial derivative norm (where ϵ is the minimum value to prevent division by zero), the adaptive damping coefficient ratio between the micro airfoil parameters and the macro layout parameters is calculated using a nonlinear decay function (such as the Sigmoid function or exponential function). Since the order of magnitude of the partial derivatives of the macroscopic layout parameters is usually much larger than that of the microscopic airfoil parameters (i.e., R≫1), the nonlinear decay function enables the microscopic airfoil parameters with smaller order of magnitude of partial derivatives to obtain a larger damping coefficient (i.e., a larger equivalent step size scaling factor) in order to balance the convergence rate of the cross-scale parameters.

[0090] The constrained optimization direction is scaled across scales according to the adaptive damping coefficient ratio to generate a parameter correction gradient with differentiated step sizes.

[0091] Specifically, based on the adaptive damping coefficient ratio and Construct the diagonal damping matrix ,in, The dimension is equal to the number of micro airfoil parameters. The identity matrix, The dimension equals the number of macro layout parameters. The identity matrix is ​​scaled across the constrained optimization direction d to generate a parameter-corrected gradient containing differentiated step sizes. .

[0092] Step S5: Based on the parameter correction gradient and the adaptive damping coefficient, perform cross-scale joint step size correction on the micro airfoil parameters and the macro layout parameters to obtain updated design parameters; repeat steps S3 to S5 based on the updated design parameters until the convergence criterion is met to obtain candidate design parameters.

[0093] Specifically, step S5 includes the following process:

[0094] Step S51: Decompose the parameter correction gradient into micro airfoil gradient and macro layout gradient, and scale them according to their respective adaptive damping coefficients to obtain micro airfoil correction step size and macro layout correction step size.

[0095] Specifically, the parameter correction gradient p is decomposed into the following blocks according to the micro airfoil parameters and macro layout parameters:

[0096]

[0097] in, For micro airfoil parameter vectors, This is a vector of macroscopic layout parameters.

[0098] p is further decomposed into micro airfoil correction step size. With the step size of macro layout adjustment .

[0099] Step S52: Based on the correction step size, the micro airfoil parameters and macro layout parameters are updated synchronously; if the updated parameters exceed the boundary of the parameter feasible region, the micro airfoil correction step size and the macro layout correction step size are reduced in the same proportion until the updated parameters fall into the feasible region to obtain the updated design parameters.

[0100] Specifically, step S52 includes the following process:

[0101] If the macro-layout parameters conform to the feasible domain boundary of the parameters after the collaborative reduction, and the current performance deviation is still greater than the preset convergence threshold, then it is determined that the macro-layout optimization space is exhausted.

[0102] Specifically, let the initial step size reduction factor be... =1, maximum number of reductions The attenuation ratio ρ∈(0,1) (e.g., ρ=0.5); perform the following for k=0,1,2,…,Kmax:

[0103] Calculate the parameter update amount for the current attempt: , ,in, and These represent the micro-airfoil correction step size and the macro-layout correction step size, respectively, which are numerically equivalent to the gradient sub-vector after damping scaling. , ; This represents the step size reduction factor in the k-th reduction iteration. ;

[0104] Calculate the updated parameters: , ,in, and Let represent the micro airfoil parameters and macro layout parameters after the k-th reduction iteration attempt, respectively.

[0105] Using the feasible region boundary Ω of the parameter to determine Is it true? If it is true, then let... If the condition is not met, then break out of the loop; if not, then let... Then, continue to the next iteration.

[0106] If in If no feasible step size is found within the next iteration, then take... And output a warning;

[0107] The final updated design parameters are as follows: , ,in, and These represent the final, updated micro airfoil parameters and macro layout parameters, respectively. This represents the final step reduction factor that ensures the updated parameters fall within the feasible region.

[0108] There exists a certain macroscopic parameter. satisfy:

[0109] or

[0110] in, and Let these represent the lower and upper bounds of the feasible region for the i-th macroscopic layout parameter, respectively. This is represented as the boundary being close to the threshold (e.g., taking 10 of the range of this dimension's parameters). −3 ), and performance deviation Greater than the convergence threshold If so, it is determined that the space for macro-layout optimization has been exhausted.

[0111] Freeze the macroscopic layout parameters, remove the same proportional collaborative reduction constraint, and perform independent iterative updates on the micro airfoil parameters based only on the micro airfoil correction step size until the updated parameters fall within the feasible region.

[0112] Specifically, let In subsequent iterations, this process remains unchanged, with step-size search and updates performed only on the micro airfoil parameters:

[0113] The step size reduction factor of micro airfoils is determined by backtracking search. : until Furthermore, it satisfies Armijo or other sufficient descent conditions; during the macro-freeze phase, no further coordinated reduction is applied to the macro-layout parameters, and performance is further improved only through independent iteration of the micro-airfoil parameters.

[0114] Step S53: When the aerodynamic performance deviation of the micro airfoil profile and the aerodynamic performance deviation of the macro layout of the whole aircraft simultaneously meet their respective preset tolerances, convergence is determined; if the performance deviation does not decrease in a set number of consecutive iterations, an early stop mechanism is triggered, and the historical optimal parameters are output as the candidate design parameters.

[0115] Specifically, let the performance deviation of the micro airfoil profile be at the t-th iteration. (For example, taking the key profile) The mean square error between the distribution and the target profile), and the overall macroscopic performance deviation of the machine are... Given micro tolerance Macro tolerance ,when and If both conditions are met, determine that the iteration has converged and output the current parameters. As candidate design parameters. If the performance deviation is within P consecutive iterations (e.g., P=5), None of them fell below the historical best value. The situation is as follows: This triggers the early stopping mechanism, halting iteration and reverting historical iterations. Minimum parameter vector Output as candidate design parameters.

[0116] Step S6: Based on the candidate design parameters of each group and their corresponding differentiated preference configurations, a comprehensive decision is made and the best option is selected to obtain the optimal aircraft layout design scheme.

[0117] Specifically, step S6 includes the following process:

[0118] Step S61: Obtain the aerodynamic performance deviation and geometric constraint violation degree corresponding to each group of candidate design parameters;

[0119] Specifically, for the i-th group of candidate design parameters Obtain aerodynamic performance deviation Degree of violation of set constraints: ,in, The degree of violation of the j-th geometric constraint (0 when satisfied, normalized positive value when out of bounds). This represents the total number of geometric constraints.

[0120] Step S62: Calculate the hard constraint penalty factor based on the geometric constraint violation degree;

[0121] Specifically, a quadratic penalty function is used to construct the hard constraint penalty term. For the i-th group of candidate parameters, the total constraint violation degree is calculated. This leads to the construction of a hard constraint penalty factor: ,in, The hard constraint penalty intensity coefficient in the decision-making stage (e.g., taking...) =10 3 ~10 6 ).

[0122] Step S63: Based on the weight vector corresponding to the differentiated preference configuration, nonlinear weighting is applied to the aerodynamic performance deviation to calculate the preference fit.

[0123] Specifically, let the differentiated preference configuration corresponding to the i-th group of candidate parameters be a normalized weight vector:

[0124]

[0125] Introducing Softmax mapping: ,in, This represents the weight vector after the Softmax nonlinear mapping. For nonlinear gain coefficients (e.g., take...) This is used to enhance the discriminative power of weight differences. Based on non-linear weights. Weighting of aerodynamic performance deviations: ,in This represents the nonlinear weighted aerodynamic performance deviation of the i-th candidate parameter group. The preference fit is calculated as follows: When the weighted bias is smaller, The closer the value is to 1, the better it matches the group's preference configuration; the greater the deviation, the better. Approaching 0.

[0126] Step S64: Combine the preference fit with the hard constraint penalty factor to calculate the overall fitness.

[0127] Specifically, the comprehensive fitness is constructed using an augmented objective form of "preference fit - constraint penalty": Preference matching item Larger is better, encouraging performance to approach the target and conform to preferences; penalty term Violations of geometric constraints are strongly suppressed to ensure the engineering feasibility of the optimal solution.

[0128] Step S65: Select the candidate design parameters with the best overall adaptability as the optimal aircraft layout design scheme.

[0129] Specifically, the overall fitness is calculated for all N sets of candidate parameters. Sort by size from largest to smallest and select the candidate design parameters with the highest overall fitness. The output is the optimal aircraft layout design scheme, namely: .

[0130] If multiple candidate fitness values ​​exist, and all are the maximum value, then the total geometric constraint violation value is further selected from these optimal solutions. The smallest possible solution is selected as the final optimal layout design to ensure the absolute feasibility of the project.

[0131] In a specific embodiment: Generative design of multi-scale aerodynamic layout for a high-altitude long-endurance flying-wing UAV.

[0132] Step S1: Obtain the structured design objectives and the feasible domain boundary of parameters.

[0133] The input natural language requirement is: "Design a high-altitude, long-endurance flying-wing unmanned aerial vehicle (UAV) with a cruising Mach number of 0.4 and a cruising altitude of 12,000 meters." The requirement analysis model calls upon the aerospace knowledge base to generate a structured set of design target parameters: cruising Mach number Ma = 0.4, target lift coefficient... =0.40, target overall lift-to-drag ratio =28.9, target pitch static stability derivative =−0.005 (takes the median of the range [-0.02, 0.01]).

[0134] Establishing a multi-scale design parameter vector: micro airfoil parameters The fourth-order CST weighting coefficients and the maximum relative thickness (6 parameters in total); Macro layout parameters The parameters are aspect ratio AR, root-to-tip ratio λ, and sweep angle Λ (three parameters in total). The feasible domain boundary Ω is set as follows: for example, aspect ratio AR∈[6,12], relative thickness... ∈[0.10,0.18], and must satisfy the cross-scale coupling constraint of wingtip thickness ≥0.06.

[0135] Step S2: Instantiate the differential preferences in parallel and generate initial design parameters.

[0136] Four sets of differentiated preference examples are defined: Process 1 (high lift priority), Process 2 (low drag priority), Process 3 (stability priority), and Process 4 (comprehensive balance). Taking Process 4 as an example, its preference weight vector is set as follows: =[0.3,0.4,0.3].

[0137] The initial parameters for process 4 are generated based on the conditional variational autoencoder (CVAE) parameter generation model. It is assumed that the initial aspect ratio AR = 12.5 (out of bounds), resulting in a wingtip thickness of only 0.05 (violating coupling constraints). A neighborhood mapping correction algorithm is executed: the AR single-dimensional projection is truncated to 12, and the thickness distribution is cooperatively scaled along the constraint normal, ultimately yielding compliant initial parameters. , where AR=11.5 and wingtip thickness 0.061.

[0138] Step S3: Aerodynamic performance simulation verification.

[0139] based on A full-aircraft aerodynamic simulation was performed using the vortex lattice method solver (Mach number 0.4 < 0.5, second-order KAMAN-Cheng correction enabled). Simulation feature data was extracted: Current... =0.38, K=25.2, =−0.012. All geometric constraints are satisfied; the degree vector is violated. ≈0.

[0140] Step S4: Calculate the sensitivity matrix and parameter correction gradient.

[0141] The partial derivatives are solved using the discrete adjoint method to construct the cross-scale sensitivity matrix G. Performance deviation is then calculated. Extracting the gradient norm: the norm of the partial derivatives of macroscopic parameters. =50.5, norm of partial derivatives of microscopic parameters =0.6, ratio R=84.2. The macroscopic damping coefficient is calculated using an adaptive damping algorithm. =0.12, micro damping coefficient =9.8. Generate parameter-corrected gradients. This allows the microscopic CST parameters to obtain a larger equivalent step size, overcoming the update stagnation dominated by the macroscopic gradient.

[0142] Step S5: Cross-scale joint step size correction and iterative convergence.

[0143] Let the initial step size reduction factor be set. =1, attenuation ratio ρ=0.5. In a certain iteration, the updated AR touch feasible region upper bound is 12, and close to the threshold. ,but =0.05 is greater than the convergence threshold of 0.001, indicating that the macroscopic optimization space is exhausted. AR=12 is frozen, collaborative reduction is lifted, and independent step-size search is performed only on the microscopic CST parameters. After 3 microscopic independent iterations, both the microscopic profile deviation and the macroscopic overall deviation simultaneously meet the tolerance, indicating convergence. Candidate design parameters for process 4 are output. .

[0144] Step S6: Comprehensive decision-making and selection of the best option.

[0145] After optimizing the four candidate solutions, the data was extracted and input into the multi-objective integrated decision model. The result of process 4 (integrated balancing) is as follows: =0.40, K=27.68, =−0.0091, wingspan 18.86m, wing area 39.44m², aspect ratio 8.99, no geometric overrun ( ).

[0146] Calculate hard constraint penalties: .

[0147] Calculate the preference fit (nonlinear gain coefficient) =2): Process 4 based on After Softmax weighting, its weighting bias is extremely small, and the calculated result is... =0.95; while Process 2 (low resistance priority) has a slightly higher lift-to-drag ratio, it has a slight thickness deviation. =0.001), resulting in =10, overall fitness drops sharply.

[0148] Calculate overall fitness: =0.95−0=0.95, the highest among all groups. The decision model ultimately selected the parameters of process 4 as the optimal solution. This solution has a lift-to-drag ratio of 27.68, which is close to the target of 28.9. The pitch static stability derivative of -0.0091 falls perfectly within the range of [-0.02, 0.01], and the geometric parameters fully comply with engineering specifications. The OpenVSP tool was used to generate a 3D model of the aircraft, and its three views are as follows. Figure 3 As shown. The total time for this design was 2004.29 seconds, which is more than 90% shorter than the traditional manual method.

[0149] Specifically, this invention constructs a cross-scale sensitivity matrix to accurately map aerodynamic performance deviations to macroscopic layout and microscopic airfoil parameters, establishing a physical causal relationship between aerodynamic response and geometric deformation, overcoming the blind optimization direction of traditional methods. For aerodynamic characteristics characterized by high global sensitivity of macroscopic parameters and low local sensitivity of microscopic parameters, an adaptive damping mechanism is introduced to balance the multi-scale correction amplitude, ensuring synchronous coordination between the overall flow field evolution and the aerodynamic characteristics of the airfoil profile. During step-size correction, cross-scale collaborative reduction and independent microscopic iteration under macroscopic freezing are implemented based on boundary constraints, maintaining the continuous stability of geometric deformation and physical response during the optimization process and avoiding parameter overshoot and scale interference. Combining parallel exploration and comprehensive decision-making based on differentiated preferences effectively resolves the limitations of optimization under multi-objective conflicts. This method achieves accurate closed-loop correction of multi-scale geometric parameters driven by aerodynamic performance deviations, significantly improving the global convergence capability and engineering practicality of aircraft layout design.

[0150] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

[0151] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A generative design method for aircraft layout based on multi-model collaboration, characterized in that, include: Step S1: Based on the requirement analysis model, the unstructured requirements of the aircraft layout are transformed to obtain structured design objectives, and the boundaries of the structured design objectives are extracted to obtain the parameter feasible domain boundaries. Step S2: Based on the differentiated preference configuration, the structured design objectives are weighted to instantiate multiple sets of design processes in parallel; The parameter generation model generates design parameters to be verified based on the structured design objectives and the differentiated preference configuration. Based on the feasible domain boundary of the parameters, the design parameters to be verified are subjected to boundary verification and cross-boundary projection correction to obtain the initial design parameters that satisfy the feasible domain. The initial design parameters include multi-scale design parameters composed of micro airfoil parameters and macro layout parameters. Step S3: Perform aerodynamic performance simulation and geometric constraint verification based on the initial design parameters to obtain performance indicators and simulation characteristic data; Step S4: Based on the simulation feature data and the multi-scale design parameters, construct a cross-scale mapping relationship to calculate the sensitivity of performance indicators to design parameters and obtain a sensitivity matrix; Calculate the performance deviation based on the performance indicators and the structured design objectives; Based on the preset constraint proxy model, the optimization direction is determined by combining the performance deviation as the objective function with the sensitivity matrix, and the adaptive damping coefficient of each scale parameter is calculated to generate the parameter correction gradient. Step S5: Based on the parameter correction gradient and the adaptive damping coefficient, perform cross-scale joint step size correction on the micro airfoil parameters and the macro layout parameters to obtain updated design parameters; repeat steps S3 to S5 based on the updated design parameters until the convergence criterion is met to obtain candidate design parameters. Step S6: Based on the candidate design parameters of each group and their corresponding differentiated preference configurations, a comprehensive decision is made and the best option is selected to obtain the optimal aircraft layout design scheme.

2. The multi-model collaboration based aircraft layout generative design method according to claim 1, characterized in that, The process of extracting the boundaries of the structured design objective to obtain the boundary of the feasible domain includes: Based on the preset engineering constraint specifications, the boundary conditions of the structured design objective are solved, and the engineering constraint specifications are mapped to the parameter interval limits corresponding to the structured design objective to construct the parameter feasible domain boundary.

3. The multi-model collaboration based aircraft layout generative design method of claim 2, wherein, The process of step S2 includes: Step S21: Based on the preset differential preference configuration, determine the preference weight vector for lift coefficient, lift-to-drag ratio and pitch static stability derivative, and instantiate multiple independent design processes in parallel according to the preference weight vector; Step S22: Use the structured design objective and the preference weight vector as the conditional input of the parameter generation model to generate design parameters to be verified, including airfoil feature vector and wing layout vector; Step S23: Match the design parameters to be verified with the boundary of the feasible region of parameters. If there are out-of-bounds parameters that exceed the boundary of the feasible region of parameters, then project and truncate or correct the out-of-bounds parameters along the normal of the feasible region constraint hyperplane, so that all design parameters to be verified converge within the feasible region of parameters, and obtain the initial design parameters.

4. The multi-model collaboration based aircraft layout generative design method of claim 3, wherein, The process of step S23 includes: The design parameters to be verified are subjected to dimension-by-dimensional comparison and cross-scale coupling verification. If a single-dimensional parameter exceeds its independent boundary, it is truncated by projection along the normal of the feasible region constraint hyperplane, and the single-dimensional parameter is forced to the boundary extremum of its out-of-bounds direction, while the other dimensional parameters remain unchanged. If the multidimensional parameters violate the cross-scale coupling constraint, a neighborhood mapping correction is performed along the normal of the feasible region constraint hyperplane. An affine mapping matrix is ​​constructed based on the normal gradient of the hyperplane at the current boundary point, and the multidimensional parameters are collaboratively scaled or translated into the feasible region along the physical association direction determined by the affine mapping matrix.

5. The multi-model collaboration based aircraft layout generative design method of claim 4, wherein, The process of step S4 includes: Step S41: Based on the simulation feature data, solve the partial derivatives of the performance index with respect to the multi-scale design parameters to construct a cross-scale sensitivity matrix; Step S42: Calculate the performance deviation between the performance index and the structured design objective; Step S43: Apply a penalty term to the performance deviation based on the physical constraint proxy model, and determine the constrained optimization direction by combining the cross-scale sensitivity matrix; Step S44: Calculate the adaptive damping coefficient based on the magnitude difference of the partial derivatives of the parameters at each scale in the cross-scale sensitivity matrix, and generate the parameter correction gradient in combination with the constrained optimization direction.

6. The multi-model collaboration based aircraft layout generative design method of claim 5, wherein, The process of step S44 includes: The first partial derivative norm of the corresponding micro airfoil parameter and the second partial derivative norm of the corresponding macro layout parameter are extracted from the cross-scale sensitivity matrix, respectively. Based on the ratio of the first partial derivative norm to the second partial derivative norm, a nonlinear decay function is used to calculate the adaptive damping coefficient ratio between the micro airfoil parameters and the macro layout parameters, so that parameters with smaller partial derivatives can obtain larger damping coefficients. The constrained optimization direction is scaled across scales according to the adaptive damping coefficient ratio to generate a parameter correction gradient with differentiated step sizes.

7. The multi-model collaboration based aircraft layout generative design method of claim 6, wherein, The process of step S5 includes: Step S51: Decompose the parameter correction gradient into micro airfoil gradient and macro layout gradient, and scale them according to their respective adaptive damping coefficients to obtain micro airfoil correction step size and macro layout correction step size. Step S52: Based on the correction step size, the micro airfoil parameters and macro layout parameters are updated synchronously; if the updated parameters exceed the boundary of the parameter feasible region, the micro airfoil correction step size and the macro layout correction step size are reduced in the same proportion until the updated parameters fall into the feasible region to obtain the updated design parameters. Step S53: When the aerodynamic performance deviation of the micro airfoil profile and the aerodynamic performance deviation of the macro layout of the whole aircraft simultaneously meet their respective preset tolerances, convergence is determined; if the performance deviation does not decrease in a set number of consecutive iterations, an early stop mechanism is triggered, and the historical optimal parameters are output as the candidate design parameters.

8. The generative design method for aircraft layout based on multi-model collaboration according to claim 7, characterized in that, The process of step S52 includes: If the macro-layout parameters conform to the feasible domain boundary of the parameters after the collaborative reduction, and the current performance deviation is still greater than the preset convergence threshold, then it is determined that the macro-layout optimization space is exhausted. Freeze the macroscopic layout parameters, remove the same proportional collaborative reduction constraint, and perform independent iterative updates on the micro airfoil parameters based only on the micro airfoil correction step size until the updated parameters fall within the feasible region.

9. The generative design method for aircraft layout based on multi-model collaboration according to claim 8, characterized in that, The process of step S6 includes: Step S61: Obtain the aerodynamic performance deviation and geometric constraint violation degree corresponding to each group of candidate design parameters; Step S62: Calculate the hard constraint penalty factor based on the geometric constraint violation degree; Step S63: Based on the weight vector corresponding to the differentiated preference configuration, nonlinear weighting is applied to the aerodynamic performance deviation to calculate the preference fit. Step S64: Combine the preference fit with the hard constraint penalty factor to calculate the overall fitness. Step S65: Select the candidate design parameters with the best overall adaptability as the optimal aircraft layout design scheme.

Images