CN-121997454-A - PILO-GAN-based aircraft overall scheme generation type design method

CN121997454ACN 121997454 ACN121997454 ACN 121997454ACN-121997454-A

Abstract

The invention relates to the technical field of aircraft design, in particular to a PILO-GAN-based aircraft overall scheme generation type design method, which converts unstructured natural language requirements into structured design parameters; the method comprises the steps of constructing a PILO-GAN network, generating initial design parameters by taking structural parameters as condition input and combining random noise, embedding aircraft design experience rule constraint and driving physical optimization based on multi-disciplinary performance evaluation, performing multi-disciplinary physical performance verification on the initial design parameters to obtain quantitative performance indexes, constructing a gradient propagation path by adopting REINFORCE algorithm, feeding performance indexes back to a generator to realize parameter updating, and finally dynamically adjusting loss term weights of constraint and physical optimization, and performing iterative optimization until preset requirements are met. The invention realizes the end-to-end intelligent generation from the requirement to the engineering viable scheme through a physical closed-loop optimization mechanism, and remarkably improves the physical consistency and engineering credibility of the design scheme.

Inventors

MA YIYUAN
WEI WENYU
HAN ZHONGHUA

Assignees

西北工业大学

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (10)

1. A PILO-GAN based aircraft population generation design method, comprising: Step S1, obtaining aircraft design requirement information, and converting unstructured natural language requirements into structured design parameters; s2, constructing a PILO-GAN network by taking the structural design parameters as conditional inputs, and generating initial design parameters of an aircraft overall scheme by combining random noise, wherein the initial design parameters are embedded with aircraft design experience rule constraints and drive physical optimization based on multidisciplinary performance evaluation; Step S3, performing multidisciplinary physical performance verification on the initial design parameters to obtain quantitative performance indexes; S4, constructing a gradient propagation path by adopting REINFORCE algorithm, and feeding the quantized performance index gradient back to a generator to update parameters so as to obtain an updating result; And S5, dynamically adjusting the weight of the constraint and physical optimization loss item, and iteratively executing the steps S2 to S4 according to the updating result until the generated design parameters meet the preset requirements to obtain target design parameters.
2. The pilot-GAN based aircraft population generation design method of claim 1, wherein the process of step S2 comprises: According to the structural design parameters and random noise, forward propagation calculation is carried out by adopting ResNet architecture through a construction generator and integrating a self-attention mechanism, so as to obtain the initial design parameters; According to the aircraft design experience criterion, calculating the constraint violation degree of the design parameters through a parameter constraint mechanism, and adopting weighted hyperbolic tangent nonlinear scaling and batch average constraint to obtain a compliance verification result; applying a maximum takeoff weight to the initial design function through a physical optimization mechanism according to multidisciplinary performance evaluation requirements, and obtaining a physical loss value by using weighted hyperbolic tangent nonlinear optimization of empty weight and fuel consumption weight; And according to the fusion characteristics of the initial design parameters and the condition labels, performing authenticity scoring by constructing a discriminator through adopting a CNN architecture and integrating a self-attention mechanism to obtain an countermeasure training signal so as to complete PILO-GAN network construction.
3. The PILO-GAN based aircraft population generation design method of claim 2, wherein the process of forward propagation computation by build generator employing ResNet architecture and integrating self-attention mechanisms comprises: , Wherein, the The generator function is represented, the initial design parameters are output, z is random noise, y is a conditional label, FC 1 、FC 2 is full link layer, resBlocks is residual block sequence, and Attention is self-Attention mechanism.
4. A PILO-GAN based aircraft population generation design method in accordance with claim 3, wherein said process of calculating said initial design parameters via a parameter constraint mechanism comprises: , Wherein, the For the parameter constraint loss value, m is the number of training batch samples, n is the total number of constraint terms, ωk is the weight coefficient of the kth constraint, sk is a nonlinear scaling factor, and Lk (i) is the original loss value of the ith sample under the kth constraint.
5. The PILO-GAN based aircraft population generation design method of claim 4, wherein the process of applying a maximum takeoff weight to the initial design function through a physical optimization mechanism using weighted hyperbolic tangent nonlinear optimization of empty weight and fuel consumption weight to obtain a physical loss value comprises: , Wherein, the In order to be a value of the physical loss, For the physical optimization of the primary loss weights, The maximum take-off weight, the use empty weight and the fuel consumption weight are respectively corresponding losses, For a corresponding non-linear scaling factor.
6. The pilot-GAN based aircraft population generation design method of claim 5, wherein the process of step S3 comprises: And performing multidisciplinary physical performance verification on the initial design parameters by adopting a vortex lattice method to obtain normalized quantitative performance indexes, wherein the quantitative performance indexes comprise the ratio of the fuel consumption weight to the maximum take-off weight, the ratio of the using empty weight to the maximum take-off weight and the lift coefficient.
7. The pilot-GAN based aircraft population generation design method of claim 6, wherein the process of step S4 comprises: generating N independent noise samples, and performing forward propagation through a generator, and performing physical performance evaluation to obtain a quantization performance index of each sample; calculating a sample mean value as a base line according to the quantitative performance index, and performing advantage function calculation to obtain the relative performance quality of each sample; outputting the gradient of the input noise through a calculation generator according to the relative performance quality, and carrying out REINFORCE algorithm estimation to obtain the gradient of the physical performance to the noise; according to the gradient, the gradient is propagated to generator parameters through a chain law, and parameter updating is carried out in combination with the gradient of the countering loss and the constraint loss to obtain the updating result.
8. The PILO-GAN based aircraft population generation design method of claim 7, wherein the process of parameter updating in combination with the gradients of countering loss and constraint loss to obtain the updated results comprises: Calculating a generator parameter gradient through a generator jacobian matrix according to the gradient of the physical property to noise; Weighting and fusing the counter loss gradient and the constraint loss gradient according to the generator parameter gradient to obtain a total gradient; And limiting the modular length of the total gradient within a preset threshold range through gradient clipping, and updating generator parameters by adopting an Adam optimizer to obtain the updating result.
9. The pilot-GAN based aircraft population generation design method of claim 8, wherein the process of step S5 comprises: Carrying out loss value monitoring on the updated result through presetting a constraint loss threshold and a physical loss threshold to obtain a loss overrun judgment result; According to the loss overrun judgment result, sampling a loss term weight coefficient within a range of 1.0-2.0 times to obtain a target weight coefficient; and retraining the PILO-GAN network according to the target weight coefficient, and iteratively executing the steps S2 to S4 to obtain the target design parameters, thereby obtaining the overall scheme of the target aircraft.
10. The pilot-GAN based aircraft population generation design method of claim 9, wherein the retraining the pilot-GAN network according to the target weight coefficients and iteratively performing steps S2 through S4, the process of deriving the target design parameters comprises: Training using a multi-stage strategy to obtain the target design parameters, wherein, Stage 1 (0-25% iteration period) optimizing only the counter loss to enable the generator to learn the data basic distribution characteristics; stage 2 (25-50% iteration period) of linearly attenuating the counterloss weight to 0, linearly increasing the constraint loss weight to 1, and guiding the generation parameters to meet the constraint of the empirical rule; stage 3 (50-75% iteration period) that the physical loss weight is linearly increased to 1, the constraint loss weight is kept at 1, and the multidisciplinary performance of the generation scheme is optimized; And stage 4 (75-100% iteration period) of fixing the physical loss and constraint loss weights to be 1, and realizing comprehensive optimization and convergence of the generation scheme to obtain the target design parameters.

Description

PILO-GAN-based aircraft overall scheme generation type design method Technical Field The invention relates to the technical field of aircraft design, in particular to a PILO-GAN-based aircraft overall scheme generation type design method. Background The overall design of the aircraft is the most critical prospective stage in the whole life cycle of the development of the aircraft, and the core performance, life cycle cost and technical feasibility of the aircraft are directly determined. The current aircraft overall scheme design still highly depends on expert experience, the design process needs repeated iteration verification, and the problems of long iteration period, difficult guarantee of the engineering effectiveness of the initial scheme and the like exist. With the increased competition and the diversification of mission requirements, aircraft designs are being transformed into unusual layouts, new propulsion systems, and intelligent design paradigms. The traditional design method is developed based on experience knowledge and iterative verification, is difficult to cope with complex nonlinear relations of multidisciplinary coupling, is excessively dependent on sequential trial and error, is limited in design space exploration, and cannot meet the requirements of high-efficiency intelligent design of modern aircrafts. The generation of artificial intelligence technology, and in particular generation of a countermeasure network (GAN), provides a new solution for aircraft design. The existing GAN related technology is applied to the fields of aerodynamic shape design and the like, but has obvious defects in the generation of an aircraft overall scheme, namely, the existing GAN related technology lacks effective fusion of inherent physical constraints coupled with multiple disciplines of the aircraft, the generation scheme is difficult to meet engineering practice requirements, the existing GAN related technology is focused on a two-dimensional wing profile or a point cloud model, the complete aircraft overall scheme is difficult to cover, the adaptability to dynamic design requirements is insufficient, and the generalization capability is weak under a small number of high-quality sample scenes, so that the aviation field knowledge and physical mechanism cannot be effectively integrated. Therefore, an aircraft overall scheme generating method capable of integrating domain knowledge and physical constraint, supporting requirement driving generation and considering scheme innovation and engineering effectiveness is needed, so as to solve the problems that the prior art depends on expert experience, has long iteration period, and has poor physical consistency. Disclosure of Invention Accordingly, the present invention provides a method for generating an aircraft overall scheme based on PILO-GAN to solve the foregoing problems in the prior art. In order to achieve the above object, the present invention provides a method for designing a general scheme generation formula of an aircraft based on PILO-GAN, comprising: Step S1, obtaining aircraft design requirement information, and converting unstructured natural language requirements into structured design parameters; s2, constructing a PILO-GAN network by taking the structural design parameters as conditional inputs, and generating initial design parameters of an aircraft overall scheme by combining random noise, wherein the initial design parameters are embedded with aircraft design experience rule constraints and drive physical optimization based on multidisciplinary performance evaluation; Step S3, performing multidisciplinary physical performance verification on the initial design parameters to obtain quantitative performance indexes; S4, constructing a gradient propagation path by adopting REINFORCE algorithm, and feeding the quantized performance index gradient back to a generator to update parameters so as to obtain an updating result; And S5, dynamically adjusting the weight of the constraint and physical optimization loss item, and iteratively executing the steps S2 to S4 according to the updating result until the generated design parameters meet the preset requirements to obtain target design parameters. Further, the process of step S2 includes: According to the structural design parameters and random noise, forward propagation calculation is carried out by adopting ResNet architecture through a construction generator and integrating a self-attention mechanism, so as to obtain the initial design parameters; According to the aircraft design experience criterion, calculating the constraint violation degree of the design parameters through a parameter constraint mechanism, and adopting weighted hyperbolic tangent nonlinear scaling and batch average constraint to obtain a compliance verification result; applying a maximum takeoff weight to the initial design function through a physical optimization mechanism according to multidisciplinary performance e