Search

CN-121980669-A - Method for rapidly predicting performance of fixed plane shape waverider based on fusion modeling

CN121980669ACN 121980669 ACN121980669 ACN 121980669ACN-121980669-A

Abstract

The invention provides a method for rapidly predicting the performance of a fixed plane shape waverider based on fusion modeling, which comprises the steps of generating the appearance of a waverider with various different design parameter combinations through a fixed plane shape waverider design method, rapidly calculating low-precision aerodynamic data of a large number of appearances through an engineering algorithm to construct a basic database, selecting a small number of typical appearances to obtain high-precision data through CFD (computational fluid dynamics) or experiments, aligning and fusing the high-precision data and the low-precision data to prepare a training sample set, inputting the training sample set into a deep neural network for training, and constructing a aerodynamic characteristic prediction model. The user only needs to input the designed Mach number, the sweepback angle, the shock wave angle and the attack angle of the target waverider, and the model can rapidly and accurately output key performance parameters such as lift-drag ratio, lift-drag coefficient, pressing core position and the like in millisecond time. The method effectively solves the problem that the traditional method is difficult to consider both in precision and efficiency, remarkably reduces the high-precision data requirement and the calculation cost, and greatly improves the design optimization efficiency of the waverider.

Inventors

  • Meng Xufei
  • LIU CHUANZHEN
  • BAI PENG
  • LI PENGFEI
  • LIU ZHOU

Assignees

  • 中国航天空气动力技术研究院

Dates

Publication Date
20260505
Application Date
20251215

Claims (10)

  1. 1. A method for rapidly predicting the performance of a fixed plane shape waverider based on fusion modeling is characterized by comprising the following steps: S1, generating a wave-taking body shape, namely determining a design state parameter, a sweepback angle parameter range and a design shock angle parameter range based on a fixed plane shape wave-taking body design method, and generating a plurality of groups of wave-taking body shape models with different design states and sweepback angles, wherein the design state parameter at least comprises a design Mach number and a flight height; s2, constructing a low-precision aerodynamic database, namely calculating aerodynamic data of each group of wave multiplier appearance models generated in the step S1 under different attack angles by adopting an engineering algorithm, wherein the aerodynamic data at least comprises a lift coefficient, a resistance coefficient and a pitching moment coefficient, and integrating the aerodynamic data to obtain the low-precision aerodynamic database; s3, obtaining high-precision aerodynamic data, namely selecting a typical appearance model covering a design parameter range from the wave-taking body appearance model generated in the step S1, and obtaining the high-precision aerodynamic data of the typical appearance model under a corresponding attack angle by adopting a CFD method or combining a wind tunnel test, wherein the high-precision aerodynamic data is consistent with the parameter types of the low-precision aerodynamic data in the step S2; s4, training sample preparation, namely carrying out data alignment and normalization processing on the low-precision aerodynamic data in the step S2 and the high-precision aerodynamic data in the step S3 to obtain a training sample set of the neural network, wherein the input characteristics of each sample in the training sample set comprise design Mach number, sweepback angle, design shock angle and attack angle, and the output label is corresponding aerodynamic parameters; s5, training the multisource fusion deep neural network, namely constructing the deep neural network, and training the neural network by using the training sample set in the step S4 to obtain a wave-taking body aerodynamic characteristic prediction model; S6, fast predicting aerodynamic characteristics, namely inputting the design Mach number, the sweepback angle, the design shock wave angle and the attack angle to be predicted of the target waverider into the prediction model obtained in the step S5, and outputting the rise resistance characteristic parameter and the stability parameter of the target waverider by the model.
  2. 2. The method for rapidly predicting the performance of a planar-shaped waverider based on fusion modeling according to claim 1, wherein the planar-shaped waverider design method in step S1 is a design method based on an oscillometric method, and the relation between a design curve and a projected planar-shaped contour line in a top view direction is deduced according to a design mach number and a shock angle by giving a value of a backswept angle of the waverider, so as to obtain the waverider appearance with a given front edge line.
  3. 3. The method for rapidly predicting the performance of the planar-shaped waverider based on the fusion modeling according to claim 1, wherein the engineering algorithm in the step S2 is at least one selected from Newton flow theory, a modified wave drag formula and a tangential wedge method, and the attack angle is in a value range of 0-20 degrees, and the step length is not more than 2 degrees.
  4. 4. The method for rapidly predicting the performance of the planar-shaped waverider based on the fusion modeling according to claim 1, wherein in the step S3, the selection number of the typical appearance models is not less than 10% of the total number of the appearance models in the step S1, and parameters of a boundary value and an intermediate value of sweepback angles of 30-70 degrees and design Mach numbers of 5-8 are covered, a solver adopted by the CFD method is a finite volume method solver based on a Reynolds average Navier-Stokes equation, and a turbulence model is selected from k-omega SST.
  5. 5. The method for fast predicting the performance of a planar-shaped waverider based on fusion modeling according to claim 1 or 3, wherein the engineering algorithm in step S2 adopts a combination of a newton flow theory and a modified wave drag formula, wherein the newton flow theory is used for calculating the axial force CA, the normal force CN and the pitching moment coefficient CMZ, and the modified wave drag formula is used for compensating the viscous drag effect.
  6. 6. The method for rapidly predicting the performance of the planar-shaped waverider based on fusion modeling as set forth in claim 1, wherein the data alignment in the step S4 is specifically that for the same appearance model and the same attack angle, low-precision aerodynamic data and high-precision aerodynamic data are in one-to-one correspondence to form a low-precision high-precision data pair, and the normalization process adopts a min-max normalization method to map the values of the input features and the output labels to a [0,1] interval.
  7. 7. The method for rapidly predicting the performance of the fixed plane shape waverider based on the fusion modeling of claim 1 is characterized in that in the step S5, the network structure of the deep neural network is that 4 nodes of an input layer are designed corresponding to Mach numbers, first sweepback angles, second sweepback angles and attack angles, 2 to 4 layers of hidden layers are adopted, the number of nodes of each layer is 16 to 24, 3 nodes of an output layer are adopted corresponding to axial force CA, normal force CN and pitching moment coefficient CMZ, a ReLU function is adopted as an activation function of the hidden layer, and a linear activation function is adopted as an output layer.
  8. 8. The method for rapidly predicting the performance of a planar shaped waverider based on fusion modeling according to claim 1 or 7, wherein in step S5, an Adam optimizer is adopted in the training process of the neural network, the initial learning rate is 0.001, and a learning rate attenuation strategy is set, wherein a re-weighted loss function is used in the training, and the loss function is: ; in the formula, l represents the number of high-precision test data samples, q represents the number of low-precision calculation data samples, ρ represents a weight coefficient, the function balances the contribution of high-precision data and low-precision data through differential weights, the model is prevented from excessively depending on the high-precision data, and meanwhile, when multiple different-precision data exist, the multi-precision data can be balanced by increasing function terms.
  9. 9. The method for fast predicting the performance of a planar shaped waverider based on fusion modeling according to claim 1 or 7, wherein in step S5, the training of the neural network adopts a fusion modeling strategy combining pre-training with fine tuning: Training the neural network only by using low-precision aerodynamic data until the loss of the verification set converges; and in the fine tuning stage, freezing parameters of part of hidden layers in the pre-training model, and updating and training the parameters of the rest layers of the network by using only high-precision aerodynamic data.
  10. 10. The method for rapidly predicting the performance of the planar-shaped waverider based on the fusion modeling according to claim 1, wherein in the step S6, the lift-drag characteristic parameters comprise lift-drag ratio, lift coefficient and drag coefficient, the stability parameters comprise a press center position, and the single prediction time of the rapid modeling is less than or equal to 0.1 second.

Description

Method for rapidly predicting performance of fixed plane shape waverider based on fusion modeling Technical Field The invention relates to the technical field of aerodynamic design of aerospace aircrafts, in particular to a method for rapidly predicting the performance of a plane-shaped waverider based on fusion modeling. Background The waverider has high lift-drag ratio in hypersonic speed stage (Mach number is greater than or equal to 4), so that the waverider becomes the key aerodynamic layout of hypersonic aircraft. The design method of the fixed plane shape waverider can generate the waverider appearance meeting the requirements of different tasks by regulating and controlling parameters such as plane shape, design shock wave angle, design Mach number and the like. However, in the performance verification step of the design method, a means for rapidly and accurately evaluating aerodynamic characteristics (including lift-drag characteristics and stability) of different shapes is needed to efficiently screen the optimal design scheme. In the prior art, an engineering algorithm (such as Newton flow theory) has high calculation speed and low precision, cannot capture complex flow field effects, a computational fluid mechanical method has high precision, extremely long calculation time and cannot meet the requirement of multi-scheme quick screening, and a traditional data driving modeling method can realize quick prediction, but severely relies on a large amount of high-precision data, has high acquisition cost and causes poor model generalization capability. In addition, the existing multisource fusion modeling method is not customized for the design characteristics of the planar waverider, a large amount of low-precision engineering data and a small amount of high-precision data generated in the design process cannot be utilized efficiently, and improvement of design efficiency is restricted. Therefore, there is an urgent need in the art for a fast prediction method for the performance of a planar-shaped waverider, which can achieve both the prediction speed and the accuracy, while significantly reducing the requirement for high-accuracy data. Disclosure of Invention The invention aims to provide a method for rapidly predicting the performance of a fixed plane shape waverider based on fusion modeling, which trains a deep neural network model by fusing low-precision engineering algorithm data and high-precision CFD/test data, realizes the millisecond-level prediction speed on the premise of ensuring the CFD level precision, and greatly reduces the dependence on the high-precision data. According to one object of the invention, the invention provides a method for rapidly predicting the performance of a fixed plane shape waverider based on fusion modeling, which comprises the following steps: S1, generating a wave-taking body shape, namely determining a design state parameter, a sweepback angle parameter range and a design shock angle parameter range based on a fixed plane shape wave-taking body design method, and generating a plurality of groups of wave-taking body shape models with different design states and sweepback angles, wherein the design state parameter at least comprises a design Mach number and a flight height; s2, constructing a low-precision aerodynamic database, namely calculating aerodynamic data of each group of wave multiplier appearance models generated in the step S1 under different attack angles by adopting an engineering algorithm, wherein the aerodynamic data at least comprises a lift coefficient, a resistance coefficient and a pitching moment coefficient, and integrating the aerodynamic data to obtain the low-precision aerodynamic database; s3, obtaining high-precision aerodynamic data, namely selecting a typical appearance model covering a design parameter range from the wave-taking body appearance model generated in the step S1, and obtaining the high-precision aerodynamic data of the typical appearance model under a corresponding attack angle by adopting a CFD method or combining a wind tunnel test, wherein the high-precision aerodynamic data is consistent with the parameter types of the low-precision aerodynamic data in the step S2; s4, training sample preparation, namely carrying out data alignment and normalization processing on the low-precision aerodynamic data in the step S2 and the high-precision aerodynamic data in the step S3 to obtain a training sample set of the neural network, wherein the input characteristics of each sample in the training sample set comprise design Mach number, sweepback angle, design shock angle and attack angle, and the output label is corresponding aerodynamic parameters; s5, training the multisource fusion deep neural network, namely constructing the deep neural network, and training the neural network by using the training sample set in the step S4 to obtain a wave-taking body aerodynamic characteristic prediction model; S6, fast predicting aerod