Airfoil parameterization method based on autoencoder model

Abstract

This invention proposes an airfoil parameterization method based on an autoencoder model. It constructs an airfoil autoencoder model that does not rely on any prior human knowledge, is entirely based on the Transformer architecture, and simultaneously generates the x and y coordinates of the airfoil. It explores a set of potential representations that can express the shape of the airfoil entirely through deep learning itself, and reconstructs the airfoil point cloud data based on these potential representations, thereby improving the applicability and parameterization accuracy of the airfoil parameterization model.

Description

Technical Field

[0001] This invention relates to the field of airfoil design, specifically to an airfoil parameterization method based on an autoencoder model. Background Technology

[0002] Parametric design is an important method in aircraft shape design. This technique reduces the dimensionality of aircraft or wing point cloud data to a small number of variables and controls these variables to control and deform the wing shape. The aircraft shape optimization design process targets these variables. The Comparative Stamping (CST) parametric method is currently a commonly used parametric method. The basic principle of CST parametric design is to linearly superimpose different curve functions within a defined design space to obtain the designed airfoil shape. Due to the high precision requirements of aircraft shape design, a key performance indicator of parametric methods is the shape reconstruction error. However, with the development of airfoil research, the design curves of new airfoil shapes are becoming increasingly complex, exceeding the design space of the CST parametric method, making it difficult to use CST parametric design. Furthermore, the design space in the CST parametric method itself is still a manually designed curve function, which introduces inductive bias, thus reducing the applicability of parametric methods to point cloud data of different shapes.

[0003] Currently, there are two main technical routes for airfoil parameterization based on deep learning technology: one is based on variational autoencoder (VAE) models, and the other is based on generative adversarial networks (GAN) models.

[0004] The first technical approach was put into practice earliest. However, previous work only parameterized the y-coordinate of the airfoil, while fixing the x-coordinate. This required that the x-coordinate of the coordinate points in the airfoil point cloud data be fixed to a set number of values, thus limiting the design space of the airfoil.

[0005] In the second technical approach, current research proposes adding a Bézier layer at the end of the parametric model to generate smooth airfoil point cloud coordinates. Subsequent developments have led to the improved BSplineGAN, which, compared to the original model, provides better shape control with fewer control parameters. While both models can reconstruct smooth airfoil data, the use of the Bézier layer essentially introduces a manually designed function, which is equivalent to introducing inductive bias.

[0006] In addition, some studies have combined these two technical approaches to build a parametric model based on VAEGAN. It does not rely on existing design formulas, but it still only models the y-coordinate of the airfoil and fixes the x-coordinate, which limits the application scope of the parametric model. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention proposes an airfoil parameterization method based on an autoencoder model. It constructs an airfoil autoencoder model that does not rely on any prior human knowledge, is entirely based on the Transformer architecture, and simultaneously generates the x and y coordinates of the airfoil. It explores a set of features that can express the shape of the airfoil entirely through deep learning itself, and can reconstruct the airfoil coordinate data based on these features, thereby improving the applicability and parameterization accuracy of the airfoil parameterization model.

[0008] The technical solution of this invention is as follows:

[0009] An airfoil parameterization method based on an autoencoder model includes the following steps:

[0010] Step 1: Adjust the data structure of the existing initial airfoil point cloud dataset to keep the data structure of the initial airfoil point cloud data consistent, and perturb the initial airfoil point cloud data to obtain new airfoil point cloud data, which together with the initial airfoil point cloud data form a multi-source airfoil point cloud dataset.

[0011] Step 2: Construct an encoder that can perform dimensionality reduction on airfoil point cloud data to obtain the potential representation of the airfoil point cloud data;

[0012] The encoder includes a serialization processing module, stacked ViT (Visual Transformer) modules, and a convolution module;

[0013] The serialization processing module serializes the input airfoil point cloud data to obtain serialized data. The serialized data is used as input to a stacked ViT module, which extracts features from the airfoil point cloud data. The features are then input to a convolution module for dimensionality reduction to obtain the latent representation of the airfoil point cloud data.

[0014] Step 3: Construct a decoder that can reconstruct the latent representation to obtain airfoil point cloud data;

[0015] The decoder includes a transposed convolution module, a stacked ViT module, a deserialization (Unpatchify) module, and a Softmax module;

[0016] The transposed convolution module upsamples the input latent representation, then extracts the serialized feature data through stacked ViT modules, and uses the deserialization (Unpatchify) module to map the serialized feature data back to the original airfoil point cloud data structure; finally, the Softmax module reconstructs the mapped data into airfoil point cloud data.

[0017] Step 4: Establish an airfoil autoencoder based on the encoder obtained in Step 2 and the decoder obtained in Step 3; the input of the airfoil autoencoder is airfoil point cloud data, and the output of the autoencoder is reconstructed airfoil point cloud data; train the airfoil autoencoder using a multi-source airfoil point cloud dataset to obtain a trained airfoil autoencoder.

[0018] Step 5: Parameterize the given airfoil point cloud data based on the airfoil autoencoder obtained in Step 4 to obtain the potential representation of the airfoil point cloud data.

[0019] Furthermore, based on the given latent representation, the corresponding airfoil point cloud data is generated using the airfoil autoencoder obtained in step 4.

[0020] Furthermore, in step 1, maintaining the consistency of the data structure means that the total number of coordinate points used to represent the airfoil in the airfoil point cloud data is the same.

[0021] Furthermore, the specific method for adjusting the data structure in step 1 is as follows: set the number of coordinate points of the shaped airfoil to N. For the initial airfoil point cloud dataset, if the number of coordinate points of a certain airfoil point cloud data is less than N, then the number of coordinate points of the airfoil point cloud data is expanded to N by interpolation. If the number of coordinate points of a certain airfoil point cloud data is greater than N, then the number of coordinate points of the airfoil point cloud data is reduced to N by sampling.

[0022] Furthermore, the method for perturbing the initial airfoil point cloud data in step 1 is as follows: for a certain initial airfoil point cloud data, the traditional parameterization method is used to parameterize it, and then the parameterized airfoil design parameters are perturbed within the set perturbation range.

[0023] Furthermore, the traditional parameterization methods include the CST parameterization method and the B-Spline parameterization method.

[0024] Furthermore, in step 2, the serialization processing module serializes the input airfoil point cloud data to obtain a serialized data structure that is a three-dimensional array with three dimensions: channel, length, and width. The channel dimension has a size C = 1, the length dimension has a size H that is the number of coordinate points after shaping in step 1, and the width dimension has a size W = 2, indicating that each coordinate point has two coordinate values, x and y.

[0025] Furthermore, the stacked ViT module is composed of several ViT Blocks connected in sequence, and each ViT Block is composed of a multi-head attention module and an FFN module connected in sequence.

[0026] Furthermore, the Softmax module in step 3 performs the following operations: First, it separates the channel data representing the x-coordinate from the channel data representing the y-coordinate in the data output by the deserialization module; then, it performs a one-dimensional convolution operation on the channel data representing the x-coordinate; next, it divides the data after the one-dimensional convolution operation into two parts with an equal number of data points along the x-coordinate dimension, and performs a Softmax operation on each part to obtain the x-coordinate spacing between adjacent data points; based on the x-coordinate spacing between adjacent data points, it calculates the x-coordinate of the reconstructed data points, and combines it with the channel data representing the y-coordinate in the data output by the deserialization module to obtain the reconstructed airfoil point cloud data.

[0027] Furthermore, the loss function for training the airfoil autoencoder in step 4 is as follows:

[0028]

[0029] Where z is the input airfoil point cloud data, For the reconstructed airfoil point cloud data, This represents the mean square error between the input airfoil point cloud data and the reconstructed airfoil point cloud data; λ is the set weight. The feature quantity used to characterize the smoothness of the reconstructed airfoil point cloud data.

[0030] Beneficial effects

[0031] This invention proposes an airfoil parameterization method based on an autoencoder model. It constructs an airfoil autoencoder model that does not rely on any prior human knowledge, is entirely based on the Transformer architecture, and simultaneously generates the x and y coordinates of the airfoil. It explores a set of potential representations that can express the shape of the airfoil entirely through deep learning itself, and reconstructs the airfoil point cloud data based on these potential representations, thereby improving the applicability and parameterization accuracy of the airfoil parameterization model.

[0032] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0033] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0034] Figure 1 This is a flowchart of the airfoil parameterization method according to an embodiment of the present invention;

[0035] Figure 2 This is a detailed diagram illustrating the construction of the airfoil autoencoder dataset according to an embodiment of the present invention;

[0036] Figure 3This is an architectural diagram of an airfoil self-encoder according to an embodiment of the present invention;

[0037] Figure 4 This is a loss curve diagram of an airfoil self-encoder according to an embodiment of the present invention;

[0038] Figure 5 The figures show the performance of the model under different compression ratios in this embodiment of the invention. (a) Overall average MAE. (b) Distribution of MAE. Detailed Implementation

[0039] The embodiments of the present invention are described in detail below. These embodiments are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0040] This embodiment uses an airfoil parameterization method based on an autoencoder model to parameterize the airfoil point cloud data of an airfoil database, such as... Figure 1 As shown, it includes the following steps:

[0041] Step 1: Adjust the data structure of the existing initial airfoil point cloud dataset to ensure that the data structure of the initial airfoil point cloud data is consistent, and perturb the initial airfoil point cloud data to obtain new airfoil point cloud data, which together with the initial airfoil point cloud data form a multi-source airfoil point cloud dataset; the resulting multi-source airfoil point cloud dataset theoretically belongs to a mixed space composed of different feature spaces.

[0042] Maintaining consistent data structure here means that the total number of coordinate points used to represent the airfoil in the airfoil point cloud data is the same.

[0043] The specific method for adjusting the data structure of the initial airfoil point cloud data is as follows: Set the number of coordinate points representing the airfoil after shaping to N. For the initial airfoil point cloud dataset, if the number of coordinate points of a certain airfoil point cloud data is less than N, then interpolation is used to expand the number of coordinate points of that airfoil point cloud data to N. If the number of coordinate points of a certain airfoil point cloud data is greater than N, then sampling is used to reduce the number of coordinate points of that airfoil point cloud data to N. In this embodiment, the number of coordinate points representing the airfoil after shaping is set to 256, and each coordinate point has two coordinate values, x and y, so each airfoil point cloud data has 512 coordinate values.

[0044] The method for perturbing the initial airfoil point cloud data is as follows: For a given initial airfoil point cloud data, parameterization is performed using traditional parameterization methods, and then the parameterized airfoil design parameters are perturbed within a set perturbation range. For example... Figure 2As shown in this embodiment, for a certain initial airfoil point cloud data, the CST parameterization method and the B-Spline parameterization method are used for parameterization respectively. Then, the airfoil design parameters corresponding to each parameterization method are perturbed 100 times to obtain the perturbed airfoil point cloud data. To ensure that the perturbed airfoil meets the requirements, the set perturbing range is 0.005Δw0.005, where Δw is the perturbing amount. Since there are 906 airfoils in the initial airfoil point cloud dataset, the final multi-source airfoil point cloud dataset includes 182,106 airfoils after adding the perturbed airfoils to the initial airfoil point cloud dataset. Each airfoil in the multi-source airfoil dataset is represented by airfoil point cloud data (i.e., x, y coordinates).

[0045] Step 2: Construct an encoder that can perform dimensionality reduction on airfoil point cloud data to obtain the potential representation of the airfoil point cloud data;

[0046] The encoder includes a serialization processing module, stacked ViT (Visual Transformer) modules, and a convolution module;

[0047] The serialization processing module serializes the input airfoil point cloud data to obtain serialized data. The serialized data is then used as input to a stacked ViT module to extract features from the airfoil point cloud data. These features are then input to a convolution module for dimensionality reduction to obtain the latent representation of the airfoil point cloud data.

[0048] The serialization process treats the airfoil point cloud data as image data. When serializing image data, a multi-channel image is stored in a computer as a three-dimensional array with three dimensions: channels (size C), length (size H), and width (size W). The serialization method involves equally spaced segments along both the length and width dimensions. Assuming the distances along the length and width dimensions are P, the original image can be segmented into a total L = (H / P)·(W / P) = HW / P. 2 Ten sub-images, each sub-image has dimensions Then, each sub-image is further expanded into a one-dimensional array of size E = C·P. 2 In this way, the image is mapped as follows:

[0049]

[0050] The final serialized data is X = [x ＜1＞T ,x ＜2＞T ,...,x ＜L＞T The sequence has a length of L, where each element x ＜i＞Each of these is a one-dimensional array formed by unfolding the sub-images.

[0051] In this invention, the airfoil point cloud data is considered as a single-channel three-dimensional array with three dimensions: channel, length, and width. The channel dimension C = 1, the length dimension H = 256 (the number of coordinate points after shaping in step 1), and the width dimension W = 2, indicating that each coordinate point has two coordinate values ​​(x and y). Specifically, the dimensions are 1 × 256 × 2. Since the width dimension is very small, this embodiment only performs equidistant segmentation along the length dimension, which is equivalent to dividing the data every P coordinate points. Therefore, the sequence length of the airfoil data is L = 256 / P, and the embedding dimension is E = 1·2P = 2P. Because this is single-channel data and the width dimension is only 2, the embedding dimension E is actually small. To use a multi-head attention mechanism in practice, this embodiment also maps the embedding dimension to a higher hidden dimension D after serialization. Therefore, the airfoil point cloud data completes the following mapping:

[0052]

[0053] The stacked ViT module is composed of several ViT Blocks connected in sequence. Each ViT Block is composed of a multi-head attention module and an FFN module connected in sequence.

[0054] The multi-head attention module is specifically defined as follows: for the input sequence X = [x ＜1＞T ,x ＜2＞T ,...,x ＜L＞T The sequence has a length of L, where each element x ＜i＞ Each is a one-dimensional array, that is Let E be the embedding dimension of the input sequence. Therefore, an input sequence is actually a matrix of dimension L×E.

[0055] Each attention head in the multi-head attention module is a self-attention mechanism, which, for a specific input element x, performs self-attention. ＜i＞ First, it needs to be converted into a query vector q. ＜i＞ Key vector k ＜i＞ AND value vector v ＜i＞ :

[0056] q ＜i＞ =x ＜i＞ W Q

[0057] k ＜i＞ =x ＜i＞ W K

[0058] v ＜i＞ =x＜i＞ W V

[0059] in These represent the mapping matrices in the self-attention mechanism. By applying the above process to each input element, the input sequence X can be transformed into...

[0060]

[0061] Subsequently, the output of each attention head is calculated using the following formula:

[0062]

[0063] In the formula, The calculated result is called the attention score, where For scaling factor, and QK T In essence, the attention score is an inner product operation between each query vector in matrix Q and the key vector in matrix K. Therefore, the attention score is a measure of the similarity between query vectors and key vectors.

[0064] The softmax function is a normalization exponential function that maps the values ​​in any n-dimensional vector to the range [0,1]. For a vector α = [α1, α2, ..., α...], ... n ], define α′=softmax(α), then the softmax function is calculated as follows:

[0065]

[0066] As can be seen, the vector after the softmax function operation satisfies

[0067]

[0068] When performing a softmax operation on the attention score matrix, each row is processed individually, and the result is called the attention weight. Finally, the self-attention output is obtained through matrix multiplication.

[0069] In scaled dot product attention, matrices Q, K, and V have the same dimension T×d. model In multi-head attention, we map them to dimension d respectively. k ,d k ,d v This process is repeated h times, and the resulting h sets of Q, K, V matrices are then subjected to parallel scaling dot product attention operations. Each attention set is called a "head," and the output of each attention head is of dimension T×d. vThe matrix obtained by concatenating the results of all attention heads yields a matrix with dimension T×hd. v The matrix is ​​then transformed to remap the output back to the initial dimension d. model Multi-headed attention is expressed by the following formula:

[0070] MultiHead(Q,K,V)=Concat(head1,head2,...,head h W O

[0071] where head i =Attention(QW i Q ,KW i K VW i V )

[0072] As can be seen, multi-head attention actually maps the original Q, K, V matrices to different subspaces through a mapping matrix. By combining the representations learned by the attention heads in different subspaces, multi-head attention learns to comprehensively understand the input information from different perspectives, which gives the attention mechanism a more powerful capability.

[0073] The FFN module specifically consists of two fully connected layers and one non-linear activation layer. The first fully connected layer is used to map the output of the multi-head attention to dimension d. ff The second fully connected layer is used to map the output of the nonlinear activation layer back to the input dimension d. model In practice, d is generally chosen. ff =4×d model In this embodiment, the GELU function is selected as the nonlinear activation layer. The GELU function is defined as follows:

[0074] GELU(x) = x·Φ(x)

[0075] Where Φ(x) is the cumulative distribution function of the Gaussian distribution.

[0076] Step 3: Construct a decoder that can reconstruct the latent representation to obtain airfoil point cloud data;

[0077] The decoder includes a transposed convolution module, a stacked ViT module, a deserialization (Unpatchify) module, and a Softmax module;

[0078] The transposed convolution module upsamples the latent representation of the input, then extracts the serialized feature data through stacked ViT modules, and uses the deserialization (Unpatchify) module to map the serialized feature data back to the original airfoil point cloud data structure; finally, the Softmax module reconstructs the mapped data into airfoil point cloud data.

[0079] The Softmax module performs the following operations: First, it separates the channel data representing the x-coordinate from the channel data representing the y-coordinate in the data output by the deserialization module; then, it performs a one-dimensional convolution operation on the x-coordinate channel data; next, it divides the data after the one-dimensional convolution operation into two parts with an equal number of data points along the x-coordinate dimension, and performs a Softmax operation on each part to obtain the x-coordinate spacing between adjacent data points; since the x-coordinate of the data points ranges from 0 to 1, the x-coordinate of the reconstructed data points is calculated based on the x-coordinate spacing between adjacent data points, and combined with the channel data representing the y-coordinate in the data output by the deserialization module to obtain the reconstructed airfoil point cloud data.

[0080] Step 4: Establish an airfoil autoencoder based on the encoder obtained in Step 2 and the decoder obtained in Step 3; the input of the airfoil autoencoder is airfoil point cloud data, and the output of the autoencoder is reconstructed airfoil point cloud data; train the airfoil autoencoder using a multi-source airfoil point cloud dataset to obtain a trained airfoil autoencoder.

[0081] The loss function for training the airfoil autoencoder is as follows:

[0082]

[0083] Where z is the input airfoil point cloud data, For the reconstructed airfoil point cloud data, the first term of the loss function The mean square error between the input airfoil point cloud data and the reconstructed airfoil point cloud data is represented by λ; λ is the set weight; the second term of the loss function represents the smoothness of the reconstructed airfoil data. To characterize the smoothness of the reconstructed airfoil point cloud data, this embodiment uses the third-order discrete difference along the y-coordinate dimension as the feature quantity. The average of the absolute values.

[0084]

[0085] The AdamW optimizer was used for optimization, and the learning rate was fixed at 1e-5 throughout the training of all models. The airfoil autoencoder was trained for approximately 1.2 million steps with a global batch size of 1024.

[0086] Step 5: Parameterize the given airfoil point cloud data based on the airfoil autoencoder obtained in Step 4 to obtain the potential representation of the airfoil point cloud data.

[0087] Example verification:

[0088] In this embodiment, we conducted a preliminary study on the impact of latent spatial dimension on the performance of airfoil autoencoders. Referring to existing classic deep learning models, we chose f = L / l = 8 as our baseline model. We constructed three identical autoencoder models, changing only the dimension of the latent representation. For these models, we fixed c = 2 and set l to 32, 24, and 16, respectively. The loss descent curves are shown below. Figure 4 As shown. The mean absolute error of these models on the test set is as follows. Figure 5 As shown.

[0089] As can be seen from the figure, the reconstruction loss of the model gradually decreases with the increase of l. The tolerance of the airfoil wind tunnel model provided in the literature is:

[0090]

[0091] When l is 24 and 16, the MAE of some samples exceeds the tolerance level. Therefore, l = 32 is more reliable in our model configuration.

[0092] In this embodiment, we compare our results with publicly available results from other state-of-the-art related works, as shown in Table 1. It should be noted that other related works only implement models for predicting coordinate values ​​y; therefore, the evaluation metrics in the table are calculated based on the y-coordinate values. rMSE in the table represents the relative mean square error, calculated using:

[0093]

[0094] The calculation is performed, where the subscript +1 in the formula indicates that the values ​​of x and x are all added by 1. Related Model 1 comes from the paper "Physically interpretable airfoil parameterization using variational autoencoder-based generative modeling. AIAA SCITECH 2024 Forum, 2024", which uses b-spline parameterization to generate smooth airfoil curves. Our model, without relying on any manually designed features, achieves a reconstruction loss only an order of magnitude different from their results. Related Model 2 comes from the paper "Towards universalparameterization: Using variational autoencoders to parameterize airfoils. AIAASCITECH 2024 Forum, 2024", which uses fully connected layers to build a relatively simple model; our results significantly outperform theirs.

[0095] Table 1 Performance comparison of the airfoil parameterization method in this embodiment

[0096]

[0097] In summary, the airfoil parameterization method of the present invention has high design accuracy when the airfoil autoencoder simultaneously generates the airfoil x-coordinate and y-coordinate.

[0098] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention without departing from the principles and spirit of the present invention.

Claims

1. An airfoil parameterization method based on an autoencoder model, characterized in that: Includes the following steps: Step 1: Adjust the data structure of the existing initial airfoil point cloud dataset to keep the data structure of the initial airfoil point cloud data consistent, and perturb the initial airfoil point cloud data to obtain new airfoil point cloud data, which together with the initial airfoil point cloud data form a multi-source airfoil point cloud dataset. Step 2: Construct an encoder that can perform dimensionality reduction on airfoil point cloud data to obtain the potential representation of the airfoil point cloud data; The encoder includes a serialization processing module, stacked ViT modules, and a convolution module; The serialization processing module serializes the input airfoil point cloud data to obtain serialized data. The serialized data is used as input to a stacked ViT module, which extracts features from the airfoil point cloud data. The features are then input to a convolution module for dimensionality reduction to obtain the latent representation of the airfoil point cloud data. Step 3: Construct a decoder that can reconstruct the latent representation to obtain airfoil point cloud data; The decoder includes a transposed convolution module, stacked ViT modules, a deserialization module, and a Softmax module; The transposed convolution module upsamples the input latent representation, then extracts the serialized feature data through stacked ViT modules, and uses the deserialization module to map the serialized feature data back to the original airfoil point cloud data structure; finally, the Softmax module reconstructs the mapped data into airfoil point cloud data. Step 4: Establish an airfoil autoencoder based on the encoder obtained in Step 2 and the decoder obtained in Step 3; the input of the airfoil autoencoder is airfoil point cloud data, and the output of the autoencoder is reconstructed airfoil point cloud data; train the airfoil autoencoder using a multi-source airfoil point cloud dataset to obtain a trained airfoil autoencoder. Step 5: Parameterize the given airfoil point cloud data based on the airfoil autoencoder obtained in Step 4 to obtain the potential representation of the airfoil point cloud data.

2. The airfoil parameterization method based on an autoencoder model according to claim 1, characterized in that: Based on the given latent representation, the corresponding airfoil point cloud data is generated using the airfoil autoencoder obtained in step 4.

3. The airfoil parameterization method based on an autoencoder model according to claim 1, characterized in that: In step 1, the data structure must remain consistent, meaning that the total number of coordinate points used to represent the airfoil in the airfoil point cloud data is the same.

4. The airfoil parameterization method based on an autoencoder model according to claim 1, characterized in that: The specific method for adjusting the data structure in step 1 is as follows: Set the number of coordinate points representing the airfoil after shaping to N. For the initial airfoil point cloud dataset, if the number of coordinate points of a certain airfoil point cloud data is less than N, then the number of coordinate points of the airfoil point cloud data is expanded to N by interpolation. If the number of coordinate points of a certain airfoil point cloud data is greater than N, then the number of coordinate points of the airfoil point cloud data is reduced to N by sampling.

5. The airfoil parameterization method based on an autoencoder model according to claim 1, characterized in that: The method for perturbing the initial airfoil point cloud data in step 1 is as follows: For a certain initial airfoil point cloud data, the traditional parameterization method is used to parameterize it, and then the parameterized airfoil design parameters are perturbed within the set perturbation range.

6. The airfoil parameterization method based on an autoencoder model according to claim 5, characterized in that: The traditional parameterization methods include CST parameterization and B-Spline parameterization.

7. The airfoil parameterization method based on an autoencoder model according to claim 1, characterized in that: In step 2, the serialization processing module serializes the input airfoil point cloud data to obtain a serialized data structure that is a three-dimensional array with three dimensions: channel, length, and width. The channel dimension has a size C = 1, the length dimension has a size H that is the number of coordinate points after shaping in step 1, and the width dimension has a size W = 2, indicating that each coordinate point has two coordinate values, x and y.

8. The airfoil parameterization method based on an autoencoder model according to claim 1, characterized in that: The stacked ViT module is composed of several ViT Blocks connected in sequence. Each ViT Block is composed of a multi-head attention module and an FFN module connected in sequence.

9. The airfoil parameterization method based on an autoencoder model according to claim 1, characterized in that: In step 3, the Softmax module performs the following operations: First, it separates the channel data representing the x-coordinate from the channel data representing the y-coordinate in the data output by the deserialization module; then, it performs a one-dimensional convolution operation on the x-coordinate channel data; next, it divides the data after the one-dimensional convolution operation into two parts with an equal number of data points along the x-coordinate dimension, and performs a Softmax operation on each part to obtain the x-coordinate spacing between adjacent data points; based on the x-coordinate spacing between adjacent data points, it calculates the x-coordinate of the reconstructed data points, and combines it with the channel data representing the y-coordinate in the data output by the deserialization module to obtain the reconstructed airfoil point cloud data.

10. The airfoil parameterization method based on an autoencoder model according to claim 1, characterized in that: The loss function for training the airfoil autoencoder in step 4 is as follows: Where z represents the input airfoil point cloud data. For the reconstructed airfoil point cloud data, This represents the mean square error between the input airfoil point cloud data and the reconstructed airfoil point cloud data; λ is the set weight. The feature quantity used to characterize the smoothness of the reconstructed airfoil point cloud data.

Images