Aircraft aerodynamic shape optimization method, device and medium based on neural network

Through a neural network-based method, the aircraft aerodynamic appearance optimization is used to optimize the aircraft using WGAN-div and MMoE-3D models, which solves the problem of low design efficiency in complex aircraft design, and achieves fast and effective aerodynamic appearance optimization, shortens design time and improves Efficiency.

CN114676639BActive Publication Date: 2024-12-03SHANGHAI UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210344429.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-31
Publication Date
2024-12-03
Estimated Expiration
2042-03-31

AI Technical Summary

Technical Problem

The prior art is difficult to quickly and efficiently design and optimize complex aerodynamic profiles of aircraft, especially in multi-body shape designs, and the lack of fast aerodynamic prediction tools leads to inefficient design and waste of resources.

Method used

Using a neural network-based method, the aircraft shape database is constructed through the WGAN-div model, combined with the convolutional neural network and MMoE-3D model for multi-task learning, and used differential evolution algorithm to optimize the shape of the aircraft to achieve aerodynamics under the Mach number and rolling angle. Data prediction and optimization.

Benefits of technology

It realizes rapid optimization design of the aerodynamic shape of the aircraft, shortens design time, improves design efficiency, and can automatically adjust the model to obtain an optimization solution when the requirements are not met. The performance is comparable to that of traditional methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114676639B_ABST
    Figure CN114676639B_ABST
Patent Text Reader

Abstract

The present invention discloses a method, device and medium for optimizing the aerodynamic shape of an aircraft based on a neural network. Specifically, the existing aircraft shape features are learned through the WGAN-div model, and potential variables are used to stably generate diversified designs. The generated aircraft shape is intelligently parameterized through a convolutional neural network, and multi-task learning is performed through the training of the MMoE-3D model to achieve rapid prediction of the axial force CA, normal force CN, pitch moment Cm and pressure center Xcp of the Mach number (0.4-0.8, 1.2-1.6) and roll angle (2°-90°) under flight conditions. Finally, these three parts are integrated and the differential evolution algorithm (DE) is used to complete the optimization design of the aircraft shape, which effectively solves the problem of slow optimization of the aircraft aerodynamic shape.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of aircraft aerodynamic shape design, and more specifically, to an aircraft aerodynamic shape optimization method, device and medium based on a neural network. Background Art

[0002] Since the aerodynamic data of the aircraft is highly nonlinearly related to the aircraft shape parameters, the design of the aircraft shape based on aerodynamics is a very difficult and time-consuming task. Therefore, it is important to design an automatic and efficient framework to design the aerodynamic shape of the aircraft to meet various performance requirements. Early aerodynamic design relied on trial and error, which resulted in low design efficiency and required not only keen intuition and a lot of experience, but also a lot of financial and material resources. Therefore, it is difficult to obtain the true optimal solution using this method.

[0003] In the 1970s, with the continuous improvement of computer technology, aerodynamic design methods based on computational fluid dynamics (CFD) began to develop. CFD, wind tunnel tests and theoretical analysis have gradually become the main tools for aerodynamic design. Because wind tunnel tests are too expensive and theoretical analysis cannot accurately describe aerodynamic details, aerodynamic design methods based on CFD have been gradually introduced. At present, aerodynamic shape optimization design methods combined with CFD are mainly divided into two categories: non-gradient methods and gradient methods. These two methods consume a lot of time due to the lack of fast aircraft aerodynamic data prediction tools.

[0004] In order to reduce the time and cost of calculation, methods combining deep generative models with CFD have begun to emerge to speed up the aerodynamic optimization design process. Yilmaz and German used a conditional generative adversarial network model (CGAN) to learn the relationship between airfoil shapes and conditional databases. It can generate the required airfoil shape based on the specified stall condition or drag coefficient as conditional data input into the network. Du and He proposed a fast and interactive design framework that enables wing aerodynamic optimization to be completed in seconds. The framework uses a BS spline-based generative adversarial network model (BSplineGAN) for airfoil shape parameterization and combines multi-layer perceptron, recurrent neural network and multi-gate expert mixture neural network for surrogate modeling to achieve scalar (drag and lift) and vector (pressure distribution) prediction for various Mach numbers and Reynolds numbers. Lei et al. used Wasserstein generative adversarial network, genetic algorithm and deep convolutional neural network to perform reverse design based on RAE2822 airfoil, and finally further verified the design performance of the proposed multi-stage method through specific design goals.

[0005] However, the research of the above researchers is effective and limited to the optimization design of airfoils, and cannot be extended to other engineering design needs. The shape of an aircraft (multi-body) is more complex and changeable than the airfoil shape of a single component, and there is currently a lack of rapid aerodynamic prediction tools for aircraft. In the design of airfoils, researchers usually use mathematical representations, which have limited flexibility and are difficult to apply to the aerodynamic shape of aircraft. Yan et al. proposed a new optimizer and tested it on the aerodynamic shape optimization of aircraft control surfaces. The new optimizer uses reinforcement learning and transfer learning to improve performance and efficiency, however, the design of aircraft wings is limited to a single shape within a continuous range, and often falls into local optimality in the early stages of aircraft shape design.

[0006] Therefore, the present invention considers adopting a method based on a neural network optimization framework to solve the above problems. Summary of the invention

[0007] The present invention is provided to solve the above problems existing in the prior art. Therefore, a method, device and medium for optimizing the aerodynamic shape of an aircraft based on a neural network are needed to solve the problem of aerodynamic shape design of an axisymmetric aircraft.

[0008] According to a first embodiment of the present invention, there is provided a method for optimizing an aircraft aerodynamic shape based on a neural network, the method comprising:

[0009] Step S100: constructing an aircraft shape database through a WGAN-div model based on an existing aircraft shape database;

[0010] Step S200: training a convolutional neural network to complete the conversion of the aircraft shape design to shape parameters;

[0011] Step S300: training the MMoE-3D model to predict the aerodynamic data of the aircraft generated by the generator, wherein the output of the MMoE-3D model consists of two groups of models, representing a subsonic model and a supersonic model respectively;

[0012] Step S400: Initialize the population of the DE algorithm to generate the first population;

[0013] Step S500: inputting the generated first population into the generator of the WGAN-div model to obtain the aircraft shape design;

[0014] Step S600: converting the aircraft shape design into shape parameters through a convolutional neural network;

[0015] Step S700: inputting the aircraft shape parameters, coordinate information and flight conditions into the MMoE-3D model, and calculating the optimized target value through the MMoE-3D model;

[0016] Step S800: Population mutation generates a new population;

[0017] Step S900: Repeat steps S500-S800, output the optimized aircraft shape, and verify whether the optimized aircraft shape meets the requirements. If it does not meet the requirements, return to step S300 and expand the training data set to adjust the MMoE-3D model.

[0018] According to a second embodiment of the present invention, there is provided an aircraft aerodynamic shape optimization device based on a neural network, the device comprising a processor, and the processor is configured to:

[0019] Based on the existing aircraft shape database, the aircraft shape database is constructed through the WGAN-div model;

[0020] Train a convolutional neural network to convert the aircraft shape design into shape parameters;

[0021] Training the MMoE-3D model to predict the aerodynamic data of the aircraft generated by the generator, wherein the output of the MMoE-3D model consists of two groups of models, representing a subsonic model and a supersonic model respectively;

[0022] Initialize the population of the DE algorithm to generate the first population;

[0023] The generated first population is input into the generator of the WGAN-div model to obtain the aircraft shape design;

[0024] The aircraft shape design is converted into shape parameters through convolutional neural network;

[0025] Inputting the aircraft shape parameters, coordinate information and flight conditions into the MMoE-3D model, and calculating the optimized target value through the MMoE-3D model;

[0026] Population mutation generates new population;

[0027] The new population is used as the input of the generator of the WGAN-div model, the optimized aircraft shape is output, and it is verified whether the optimized aircraft shape meets the requirements. If it does not meet the requirements, the training data set is expanded to adjust the MMoE-3D model.

[0028] According to a third aspect of the present invention, a non-transitory computer-readable storage medium storing instructions is provided. When the instructions are executed by a processor, the method according to various embodiments of the present invention is performed.

[0029] According to the aircraft aerodynamic shape optimization method, device and medium based on neural network of each scheme of the present invention, the existing aircraft shape features are learned through the WGAN-div model, and the potential variables are used to stably generate diversified designs. The generated aircraft shape is intelligently parameterized through the convolutional neural network, and the multi-task learning of the training of the MMoE-3D model is used to achieve the rapid prediction of the axial force CA, normal force CN, pitch moment Cm and pressure center Xcp of the Mach number (0.4-0.8, 1.2-1.6) and roll angle (2°-90°) under flight conditions. Finally, these three parts are integrated and the differential evolution algorithm (DE) is used to complete the optimization design of the aircraft shape, which effectively solves the problem of slow optimization of the aircraft aerodynamic shape. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In the drawings, which are not necessarily drawn to scale, the same reference numerals may describe similar parts in different views. The same reference numerals with letter suffixes or different letter suffixes may represent different instances of similar parts. The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the specification and claims, are used to illustrate the embodiments of the invention. When appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0031] Figure 1 A flowchart of a method for optimizing an aircraft's aerodynamic shape based on a neural network according to an embodiment of the present invention is shown.

[0032] FIG2(A) shows a comparison of the vehicle appearance before and after optimization at subsonic speed.

[0033] FIG2(B) shows a comparison of lift-to-drag ratios before and after optimization at subsonic speed.

[0034] FIG2(C) shows a comparison of the pressure center change rate before and after optimization at subsonic speed.

[0035] FIG3(A) shows a comparison of the vehicle appearance before and after optimization at supersonic speed.

[0036] FIG3(B) shows a comparison of lift-to-drag ratios before and after optimization at supersonic speed.

[0037] FIG3(C) shows a comparison of the pressure center change rate before and after optimization at supersonic speed. DETAILED DESCRIPTION

[0038] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings and specific embodiments, but are not intended to limit the present invention. For the various steps described herein, if there is no necessity for a causal relationship between each other, the order in which they are described as examples herein should not be regarded as a limitation, and those skilled in the art should know that they can be adjusted in order, as long as the logic between them is not destroyed, resulting in the inability to implement the entire process.

[0039] Figure 1 The flowchart of a method for optimizing the aerodynamic shape of an aircraft based on a neural network according to an embodiment of the present invention is shown. The embodiment of the present invention provides a method for optimizing the aerodynamic shape of an aircraft based on a neural network. Figure 1 As shown, the method starts at step S100, and an aircraft shape database is constructed through a WGAN-div model based on an existing aircraft shape database. The "WGAN-div model" described in this article is called the Wasserstein GAN-div model, which represents a deep learning model.

[0040] In some embodiments, the aircraft shape database is constructed by using the WGAN-div model using the following formulas (1) and (2):

[0041] G=g(z,θ) Formula (1)

[0042] z~N(0,1),z∈(-3,3) Formula (2)

[0043] Wherein, G represents the aircraft shape generator of the WGAN-div model, z represents the latent vector of the truncated normal distribution, and θ represents the hyperparameters in the aircraft shape generator.

[0044] In step S200, a convolutional neural network (CNN) is trained to complete the conversion of the aircraft shape design to shape parameters.

[0045] In some embodiments, the convolutional neural network is trained by the following formula (3) to complete the conversion of the aircraft shape design to the shape parameters:

[0046]

[0047] Among them, C represents the relationship between the aircraft shape design and shape parameters, and ε represents the convolutional neural network parameters.

[0048] In step S300, the MMoE-3D model is trained to predict the aerodynamic data of the aircraft generated by the generator. The output of the MMoE-3D model consists of two groups of models (M suband M sup ), representing the subsonic model (Ma = [0.4, 0.6, 0.8]) and the supersonic model (Ma = [1.2, 1.4, 1.6]). It should be noted that the "MMoE-3D model" described in this article is a multi-task learning model.

[0049] In some embodiments, the MMoE-3D model is trained to predict the aerodynamic data of the aircraft generated by the generator using the following formula (4):

[0050]

[0051] Where M represents the potential nonlinear relationship between aerodynamic data and vehicle shape, represents the predicted value of M, a represents the coordinate information, c represents the flight conditions, and ∈ represents the convolutional neural network parameters.

[0052] In step S400, the population of the DE algorithm is initialized to generate a first population. The DE algorithm is a differential evolution algorithm.

[0053] In step S500, the generated first population is input into the generator of the WGAN-div model to obtain the aircraft shape design.

[0054] In step S600, the aircraft shape design is converted into shape parameters through a convolutional neural network;

[0055] In step S700, the aircraft shape parameters, coordinate information and flight conditions are input into the MMoE-3D model, and the optimized target value is calculated by the MMoE-3D model;

[0056] In step S800, the population mutates to generate a new population;

[0057] Finally, in step S900, steps S500-S800 are repeated to output the optimized aircraft shape, and verify whether the optimized aircraft shape meets the requirements. If it does not meet the requirements, return to step S300 and expand the training data set to adjust the MMoE-3D model. It should be noted that repeating steps S500-S800 specifically uses the new population as a new input to the generator of the WGAN-div model, and executes steps S500-S800 again. The number of times steps S500-S800 are repeated is determined according to the number of evolutions of DE, for example, it can be 50 times, 100 times, 150 times, etc. The embodiment of the present invention does not impose specific limitations on this.

[0058] In some embodiments, the following formula (5) is used to verify whether the optimized aircraft shape meets the requirements:

[0059]

[0060]

[0061] Among them, F is the objective function of optimization, C represents the relationship between the aircraft shape design and shape parameters, z represents the latent vector of the truncated normal distribution, θ represents the hyperparameters in the aircraft shape generator, and st stands for subject to, which means making...satisfy...

[0062] The following embodiment of the present invention will use a design case to illustrate the effectiveness and advancement of the method provided by the embodiment of the present invention.

[0063] Taking the optimization of the rate of change of the center of pressure and the lift-to-drag ratio (calculated by the axial and normal forces) at a roll angle of 45° as a design case, the optimization objective function F is expressed as:

[0064]

[0065] Where w1 is the weight of the lift and drag values, set to 2 / 7, w2 is the weight of the drag term, set to 1 / 7, and w3 is the weight associated with the rate of change function of the center of pressure, set to 4 / 7. For lift-to-drag ratio and drag, the sum of the values ​​is added up and averaged over the range of Mach numbers studied. For the rate of change of the center of pressure, it is the sum of the rate of change with Mach number and the rate of change with angle of attack. Design cases are performed at subsonic and supersonic speeds, and the optimization problem is shown in Table 1.

[0066] Table 1. Aircraft optimization problem

[0067] Function or variable describe optimization F Maximize F variable z Latent variables constraint -3<z<3 Constraints on latent variables condition α Angle of attack Φ Roll Angle Ma Mach number

[0068] The shape of the aircraft was optimized under the flight condition of Φ = 45°. The shape constraint method used in the optimization comparison experiment was WGAN-div+CNN. The optimization target value of the baseline shape was 2.31. The number of evolutions of DE was 100 and the population size was 100. Figure 2(A) shows the aerodynamic shape of the missile at subsonic speed before and after optimization. The aircraft was optimized by DATCOM and MMoE-3D respectively. From Figure 2(B) and Figure 2(C), it can be seen that both of them moved the front wing position forward and increased the area, while the tail wing was almost unchanged. In particular, after optimization with the MMoE-3D model, the chord length of the front wing increased more significantly, which increased the lift-to-drag ratio, but the rate of change of the center of pressure with the Mach number was slightly larger. The time for the DATCOM optimization method was 21590s, and the comprehensive performance after optimization was 4.355. The optimization method using MMoE-3D significantly shortened the time to only 2256s, and the comprehensive performance of the optimized aircraft was 4.226.

[0069] The experimental setup for supersonic speed is consistent with that for subsonic speed. The optimization target value of the baseline shape is 1.887. Figure 3(A) shows the comparison of the shape of the aircraft before and after optimization. The optimization trends of the DATCOM and MMoE-3D models are consistent. After optimization, the position of the front wing is moved back and optimized to a rectangular wing, and the tail area is reduced to a delta wing. It can be seen from Figures 3(B) and 3(C) that compared with the baseline, the center of pressure change rate of the aircraft optimized by both optimization methods is significantly reduced, while the missile lift-to-drag ratio after DATCOM optimization is larger. The comprehensive performance after DATCOM optimization is 2.575, and the comprehensive performance after MMoE-3D optimization is 2.565. The optimization capabilities of the two are almost the same. In terms of optimization speed, the supersonic optimization using the MMoE-3D model requires more time of 4152s compared to the subsonic optimization, but it is still shorter than the 21590s required by DATCOM.

[0070] Experimental results show that the framework can obtain the optimal design solution quickly, and its performance is the same as that of the traditional optimization method, which can provide certain reference and guidance for the optimal design of the aerodynamic shape of the aircraft.

[0071] The embodiment of the present invention also provides an aircraft aerodynamic shape optimization device based on a neural network, the device comprising a processor, the processor being configured to: construct an aircraft shape database through a WGAN-div model based on an existing aircraft shape database; train a convolutional neural network to complete the conversion from aircraft shape design to shape parameters; train an MMoE-3D model to predict the aerodynamic data of the aircraft generated by a generator, the output of the MMoE-3D model consisting of two groups of models, representing a subsonic model and a supersonic model respectively; initialize the population of the DE algorithm to generate a first population; and generate the first population. The population is input into the generator of the WGAN-div model to obtain the aircraft shape design; the aircraft shape design is converted into shape parameters through a convolutional neural network; the aircraft shape parameters, coordinate information and flight conditions are input into the MMoE-3D model, and the optimized target value is calculated through the MMoE-3D model; the population is mutated to generate a new population; the new population is used as the input of the generator of the WGAN-div model, the optimized aircraft shape is output, and it is verified whether the optimized aircraft shape meets the requirements. If it does not meet the requirements, the training data set is expanded to adjust the MMoE-3D model.

[0072] It should be noted that the processor may be a processing device including one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), a graphics processing unit (GPU), etc. More specifically, the processor may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor running other instruction sets, or a processor running a combination of instruction sets. The processor may also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a system on a chip (SoC), etc.

[0073] The processor may be communicatively coupled to the memory and configured to execute computer executable instructions stored thereon to perform the aircraft aerodynamic shape optimization method based on a neural network according to various embodiments of the present invention.

[0074] In some embodiments, the processor is further configured to construct an aircraft shape database based on the existing aircraft shape database by using the WGAN-div model using the following formulas (1) and (2):

[0075] G=g(z,θ) Formula (1)

[0076] z~N(0,1),z∈(-3,3) Formula (2)

[0077] Wherein, G represents the aircraft shape generator of the WGAN-div model, z represents the latent vector of the truncated normal distribution, and θ represents the hyperparameters in the aircraft shape generator.

[0078] In some embodiments, the processor is further configured to train a convolutional neural network to complete the conversion of the aircraft shape design to shape parameters by using the following formula (3):

[0079]

[0080] Among them, C represents the relationship between the aircraft shape design and shape parameters, and ε represents the convolutional neural network parameters.

[0081] In some embodiments, the processor is further configured to train the MMoE-3D model to predict the aerodynamic data of the aircraft generated by the generator using the following formula (4):

[0082]

[0083] Where M represents the potential nonlinear relationship between aerodynamic data and vehicle shape, represents the predicted value of M, a represents the coordinate information, c represents the flight conditions, and ∈ represents the convolutional neural network parameters.

[0084] In some embodiments, the processor is further configured to verify whether the optimized aircraft shape meets the requirements by using the following formula (5):

[0085]

[0086]

[0087] Wherein, F is the objective function of optimization, C represents the relationship between the aircraft shape design and the shape parameters, z represents the latent vector of the truncated normal distribution, and θ represents the hyperparameters in the aircraft shape generator.

[0088] In some embodiments, the optimization objective function F is expressed as:

[0089]

[0090] Among them, w1 is the weight of the lift and drag values, which is set to 2 / 7, w2 is the weight of the drag term, which is set to 1 / 7, and w3 is the weight associated with the pressure center change rate function, which is set to 4 / 7.

[0091] The neural network-based aircraft aerodynamic shape optimization device described in each embodiment of the present invention is consistent in effect and method, and will not be repeated here.

[0092] An embodiment of the present invention further provides a non-transitory computer-readable medium storing instructions, and when the instructions are executed by a processor, the method according to any embodiment of the present invention is executed.

[0093] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present invention with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. The elements in the claims will be interpreted broadly based on the language adopted in the claims, and are not limited to the examples described in this specification or during the implementation of this application, and the examples will be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered as examples only, and the true scope and spirit are indicated by the following claims and the full scope of their equivalents.

[0094] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the present invention. This should not be interpreted as a feature of an invention that is not claimed for protection being necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of the embodiments of a specific invention. Thus, the following claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present invention should be determined with reference to the full scope of the equivalent forms of the attached claims and these claims.

Claims

1. A method for optimizing the aerodynamic shape of an aircraft based on a neural network, characterized in that: The method comprises: Step S100: constructing an aircraft shape database using the WGAN-div model based on an existing aircraft shape database; Step S200: training a convolutional neural network to convert the aircraft shape design into shape parameters; Step S300: training the MMoE-3D model to predict the aerodynamic data of the aircraft generated by the generator, wherein the output of the MMoE-3D model consists of two groups of models, representing a subsonic model and a supersonic model respectively; Step S400: Initialize the population of the DE algorithm to generate the first population; Step S500: inputting the generated first population into the generator of the WGAN-div model to obtain the aircraft shape design; Step S600: converting the aircraft shape design into shape parameters through a convolutional neural network; Step S700: inputting aircraft shape parameters, coordinate information, and flight conditions into the MMoE-3D model, and calculating the optimized target value through the MMoE-3D model; Step S800: Population mutation generates a new population; Step S900: Repeat steps S500-S800, output the optimized aircraft shape, and verify whether the optimized aircraft shape meets the requirements. If it does not meet the requirements, return to step S300 and expand the training data set to adjust the MMoE-3D model.

2. The method according to claim 1, characterized in that Based on the existing aircraft shape database, the aircraft shape database is constructed by using the WGAN-div model using the following formulas (1) and (2): G=g(z,θ) Formula (1) z~N(0,1),z∈(-3,3) Formula (2) Wherein, G represents the aircraft shape generator of the WGAN-div model, z represents the latent vector of the truncated normal distribution, and θ represents the hyperparameter in the aircraft shape generator.

3. The method according to claim 1, characterized in that The convolutional neural network is trained using the following formula (3) to complete the conversion of aircraft shape design to shape parameters: Among them, C represents the relationship between the aircraft shape design and shape parameters, and ε represents the convolutional neural network parameters.

4. The method according to claim 1, wherein The MMoE-3D model is trained to predict the aerodynamic data of the aircraft generated by the generator using the following formula (4): Where M represents the potential nonlinear relationship between aerodynamic data and vehicle shape, represents the predicted value of M, a represents the coordinate information, c represents the flight conditions, and ∈ represents the convolutional neural network parameters.

5. The method according to any one of claims 1 to 4, characterized in that Verify whether the optimized aircraft shape meets the requirements by using the following formula (5): Where F is the objective function for optimization, C represents the relationship between the aircraft shape design and shape parameters, z represents the latent vector of the truncated normal distribution, and θ represents the hyperparameters in the aircraft shape generator.

6. The method according to claim 5, characterized in that The objective function F of the optimization is expressed as: Among them, w1 is the weight of the lift and drag values, which is set to 2 / 7, w2 is the weight of the drag term, which is set to 1 / 7, and w3 is the weight related to the center of pressure change rate function, which is set to 4 / 7.

7. A device for optimizing the aerodynamic shape of an aircraft based on a neural network, characterized in that: The apparatus comprises a processor configured to: Based on the existing aircraft shape database, the aircraft shape database is constructed through the WGAN-div model; Train a convolutional neural network to convert aircraft shape design into shape parameters; Training the MMoE-3D model to predict the aerodynamic data of the aircraft generated by the generator, wherein the output of the MMoE-3D model consists of two sets of models, representing a subsonic model and a supersonic model respectively; Initialize the population of the DE algorithm to generate the first population; The generated first population is input into the generator of the WGAN-div model to obtain the aircraft shape design; The aircraft shape design is converted into shape parameters through convolutional neural networks; Inputting aircraft shape parameters, coordinate information, and flight conditions into the MMoE-3D model, and calculating optimized target values ​​using the MMoE-3D model; Population mutation generates new population; The new population is used as the input of the generator of the WGAN-div model, and the optimized aircraft shape is output. It is then verified whether the optimized aircraft shape meets the requirements. If it does not meet the requirements, the training dataset is expanded to adjust the MMoE-3D model.

8. The device according to claim 7, characterized in that The processor is further configured to construct an aircraft shape database based on the existing aircraft shape database by using the WGAN-div model using the following formulas (1) and (2): G=g(z,θ) Formula (1) z~N(0,1),z∈(-3,3) Formula (2) Wherein, G represents the aircraft shape generator of the WGAN-div model, z represents the latent vector of the truncated normal distribution, and θ represents the hyperparameter in the aircraft shape generator.

9. The device according to claim 7, characterized in that The processor is further configured to train a convolutional neural network to complete the conversion of the aircraft shape design to shape parameters by using the following formula (3): Among them, C represents the relationship between the aircraft shape design and shape parameters, and ε represents the convolutional neural network parameters. 10 . A non-transitory computer-readable storage medium storing instructions, which, when executed by a processor, executes the method according to claim 1 .

Citation Information

Patent Citations

  • Aircraft aerodynamic characteristic prediction method based on neural network

    CN113609596A

  • Aircraft assembly line resource allocation optimization method based on neural network and genetic algorithm

    CN114004008A