Directed distance field-based aerodynamic configuration parameterization method for trans-configuration aircraft

Abstract

The invention discloses a directed distance field-based trans-configuration aircraft aerodynamic configuration parameterization method. The method comprises the following steps of 1, constructing an aircraft aerodynamic configuration database; 2, constructing and training a directed distance field neural network model; and step 3, representing the shape of the aircraft through the trained directed distance field neural network. By means of the method, the directed distance field is introduced into aerodynamic configuration parameterization of the aircraft, compared with a traditional parameterization method, the method can achieve unified expression of the cross-configuration aircraft, the aerodynamic optimization design space of the aircraft is expanded, and an advanced technical means is provided for high-quality rapid iteration of the aircraft.

Description

Technical Field

[0001] The invention relates to a method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a directed distance field, and belongs to the field of aerodynamic shape parameterization. Background Art

[0002] Aerodynamic shape parameterization plays a crucial role in aerodynamic optimization design. By transforming complex aircraft aerodynamic shapes into a finite set of parameters, designers can not only more flexibly adjust the shape to meet design requirements but also integrate it with tools such as computer-aided design and computational fluid dynamics, automating and intelligentizing the design process and minimizing manual intervention. During aircraft optimization design, the design space is defined by the aerodynamic shape parameterization method. To improve design accuracy, a large number of optimization design variables is often required. However, the increase in design variables leads to an excessively large dimensionality of the design space, dramatically increasing the cost of obtaining a surrogate model with acceptable accuracy and making the optimization algorithm's search for feasible solutions unsustainable, a problem known as the "curse of dimensionality." Therefore, it is urgent to develop a geometric parameterization method that can cover the entire design space with a minimum of design variables to ensure that the optimal solution is included, while also possessing sufficient flexibility and accuracy to accurately describe aerodynamic shape variations, in order to meet the complex and ever-changing design requirements of future aircraft.

[0003] Whether using optimization or empirically-based design, the overall aircraft design process follows a basic process: first, conceptual design is performed to propose a preliminary shape, then this shape is evaluated and analyzed, and finally, improvements are made based on simulation results, with this process iterating repeatedly. Therefore, efficiently and accurately characterizing the shapes of different aircraft and generating corresponding numerical models is crucial to improving the efficiency of the entire design process.

[0004] Currently, mainstream aerodynamic shape characterization methods can be divided into construction methods and deformation methods based on how design variables affect the shape surface. Construction methods directly characterize the aircraft shape based on the parameters of specified design variables, such as polynomial methods, B-spline curves, partial differential equations (PDEs), and CST methods. Deformation methods, on the other hand, deform existing shapes to create new ones. These methods include discrete methods, analytical methods, basis vector methods, and free-form deformation methods (FFDs).

[0005] Although the traditional aircraft shape characterization methods mentioned above have been widely used in various fields, the tasks undertaken by future aircraft are becoming increasingly complex, the design iteration cycle is getting longer and longer, and the requirements for aircraft shape design in the early and late stages of design are not the same. This puts high demands on the flexibility and freedom of aircraft parameterization methods. Existing methods are difficult to meet the growing demand for flexible design. Summary of the Invention

[0006] In response to the above problems, in order to meet the needs of flexible design of aircraft aerodynamic shape without being restricted by conventional parameterization methods, the present invention aims to provide a cross-configuration aircraft aerodynamic shape parameterization method based on a signed distance field, providing a more advanced technical means for high-quality and rapid iteration of aircraft aerodynamic shape design.

[0007] In order to achieve the purpose of the invention, the technical solution of the present invention is as follows: A method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a signed distance field comprises the following steps: Step 1: Build an aircraft aerodynamic shape database; Step 2: Build and train a directed distance field neural network model; Step 3: Characterize the aircraft shape through the trained directed distance field neural network.

[0008] In the method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a signed distance field, in step 1, the aircraft aerodynamic shape database contains aircraft of different configurations, each configuration contains the minimum number of different shape CAD models required to meet the accuracy of reconstructing the aircraft shape, and all models in the database are preprocessed to obtain a signed distance value library corresponding to each aircraft, each sample in the library consists of a set of three-dimensional space coordinate points and corresponding signed distance field values.

[0009] The method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a signed distance field is described. The different configurations are hypersonic aircraft configurations, including a waverider body, a lifting body, and a wing-body combination. Each configuration contains an initial shape, and different shapes of the same configuration are obtained by sampling the initial shape using an existing parameterization method.

[0010] The method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a signed distance field, and the preprocessing method are as follows: translating each shape until the center of mass coincides with the origin of the Cartesian coordinate system; scaling the shape to a maximum length of unit length; and then sampling the space of a cube with its center at the origin and an edge length of unit length, thereby encrypting the sampling of the position on the aircraft surface.

[0011] In the method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a signed distance field, in step 2, the model structure consists of a multilayer perceptron and a residual neural network, using dropout technology. The model inputs are latent variables representing the specific aircraft shape and a single three-dimensional spatial coordinate point, and the output is the corresponding signed distance field value. The signed distance field neural network model is trained by assigning a randomly initialized latent variable to each aircraft shape at the beginning of training. The latent variable and the neural network weights are then optimized together through backpropagation until the loss function value no longer decreases.

[0012] The method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a signed distance field, in step 3, the characterization of the aircraft shape is to input different latent variables into the signed distance field neural network model to obtain signed distance field functions of different aircraft shapes, and the corresponding aircraft shape grid can be obtained through the Marching Cube algorithm.

[0013] The advantages of the present invention compared with the prior art are: The present invention provides a method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a signed distance field, in order to improve the problem that the current traditional aircraft geometric parameterization design method has too strong constraints on the aerodynamic shape, lacks flexibility, and leads to limited design optimization space. The established characterization method enables the iterative process of shape design to be free from the restrictions of geometric parameters, and uses the learned potential characteristic parameters of the aircraft as new design variables to achieve a generalized expression of the shape, while also avoiding the dimensional disaster problem in the optimization design process. In the shape optimization design process, the flexible design of the shape can be completed by changing the characteristic parameters, avoiding the repeated iterations between multiple software in the traditional optimization design process, and achieving high-fidelity direct generation from characteristic parameters to three-dimensional shape and mesh, effectively reducing the aircraft optimization design cycle and cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The figure is a flow chart of the aerodynamic shape parameterization method of cross-configuration aircraft based on signed distance field.

[0015] Figure 2 Schematic diagram of the lifting body of the initial shape model.

[0016] Figure 3 Schematic diagram of the waverider with the initial shape model.

[0017] Figure 4 Schematic diagram of the wing-body assembly of the initial shape model.

[0018] Figure 5 This is a structural diagram of the directed distance field neural network model, where the numbers represent dimensions. DETAILED DESCRIPTION

[0019] The present invention is further described below with reference to the accompanying drawings and embodiments. Example

[0020] like Figure 1 As shown, the present invention provides a method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a signed distance field, comprising the following steps: Step 1: Build an aircraft aerodynamic shape database; Step 2: Build and train a directed distance field neural network model; Step 3: Use the trained directed distance field neural network to characterize the aircraft shape.

[0021] Wherein in step 1, the aircraft of different configurations include a lifting body, a waverider body and a wing-body combination configuration. Figure 2 、 Figure 3 、 Figure 4 The initial shape schematic diagrams of the above three configurations are shown respectively, and the three initial shapes are parameterized using the CST parameterization method. Then, the uniform design sampling method is used to sample based on the initial shapes. 400 shapes are sampled for each configuration, and a total of 1,200 shapes are used as the database.

[0022] To train the signed distance field neural network model, the shapes in the database need to be preprocessed to obtain corresponding signed distance field samples. Each sample consists of a set of three-dimensional spatial coordinate points and corresponding signed distance field values. The preprocessing method used here is as follows: each shape is translated until the center of mass coincides with the origin of the Cartesian coordinate system; the shape is scaled to a maximum length of 1 meter; and then a cube with a side length of 1 meter and a center at the origin is sampled. To obtain a more detailed aircraft shape, the positions of the aircraft surface are densely sampled, a total of 100,000 times.

[0023] In step 2, the structure of the constructed directed distance field neural network model is as follows: Figure 5 Specifically, the neural network consists of six 256-dimensional fully connected layers, with skip connections between the input layer and the fourth layer. Each layer uses dropout. Weight normalization is used for normalization, the tanh function model is used as the activation function, and the Adam optimization algorithm is used for optimization. The network training inputs are 16-dimensional latent variables representing specific aircraft shapes and a single 3D spatial coordinate point, and the output is the corresponding signed distance field value. At the beginning of training, all aircraft shapes are assigned a latent variable randomly initialized from a Gaussian distribution with mean 0 and variance 0.01. Backpropagation is then used to optimize the latent variables and the neural network weights until the loss function no longer decreases.

[0024] In step 3, the trained signed distance field neural network is used to characterize the aircraft shape. After fixing the weights of the neural network, different latent variables are input into the signed distance field neural network model to obtain the signed distance field functions of different aircraft shapes. The corresponding aircraft shape mesh is obtained through the Marching Cube algorithm. The obtained mesh can be directly input into the meshing software as the aircraft surface mesh.

[0025] This method, based on a directed distance field neural network, enables efficient parametric representation of aircraft shapes. Compared to traditional parametric techniques, this method offers the ability to uniformly represent aircraft shapes across configurations, effectively broadening the scope of aerodynamic optimization design exploration and providing an advanced solution for efficient design iteration and performance improvement of aircraft.

[0026] Although the present invention has been disclosed above in terms of preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solutions of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the scope of protection of the technical solutions of the present invention.

Claims

1. A method for parameterizing the aerodynamic shape of a cross-configuration aircraft based on a signed distance field, characterized in that: The following steps are involved: Step 1: Build an aircraft aerodynamic shape database; Step 2: Build and train a directed distance field neural network model; Step 3: Characterize the aircraft shape through the trained directed distance field neural network.

2. The method for parameterizing aerodynamic shape of a cross-configuration aircraft based on a signed distance field according to claim 1, characterized in that: In step 1, the aircraft aerodynamic shape database contains aircraft of different configurations, each configuration containing the minimum number of different shape CAD models required to reconstruct the aircraft shape accuracy. All models in the database are preprocessed to obtain a signed distance value library corresponding to each aircraft. Each sample in the library consists of a set of three-dimensional space coordinate points and corresponding signed distance field values.

3. The method for parameterizing aerodynamic shape of a cross-configuration aircraft based on a signed distance field according to claim 2, characterized in that: The different configurations are hypersonic aircraft configurations, including waverider bodies, lifting bodies and wing-body combinations, and each configuration contains an initial shape. Different shapes of the same configuration are obtained by sampling the initial shape using existing parameterization methods.

4. The method for parameterizing aerodynamic shape of a cross-configuration aircraft based on a signed distance field according to claim 2, characterized in that: The preprocessing method is as follows: translate each shape until the center of mass coincides with the origin of the Cartesian coordinate system; scale the shape to a maximum length of unit length; then sample the space of a cube with its center at the origin and its edges of unit length, and perform encrypted sampling on the position of the aircraft surface.

5. The method for parameterizing aerodynamic shape of a cross-configuration aircraft based on a signed distance field according to claim 1, characterized in that: In step 2, the model architecture consists of a multilayer perceptron and a residual neural network, using dropout technology. The model inputs are latent variables representing the specific aircraft shape and a single 3D spatial coordinate point, and the output is the corresponding signed distance field value. To train the signed distance field neural network model, at the beginning of training, all aircraft shapes are assigned a randomly initialized latent variable. Backpropagation is then used to optimize the latent variable and the neural network weights until the loss function no longer decreases.

6. The method for parameterizing aerodynamic shape of a cross-configuration aircraft based on a signed distance field according to claim 1 or 5, characterized in that: In step 3, the characterization of the aircraft shape is to input different latent variables into the signed distance field neural network model to obtain the signed distance field function of different aircraft shapes, and the corresponding aircraft shape grid can be obtained through the Marching Cube algorithm.

Images