Aerodynamic heat rapid analysis method for object plane of deformation structure aircraft
By building a training sample database on the server side and training a conditional diffusion model, the problems of large amount of calculation and low accuracy in aerodynamic thermal prediction of deformed structure aircraft are solved, and fast and high-precision aerodynamic thermal prediction is achieved.
Patent Information
- Application Number
- CN202411861055.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-17
- Publication Date
- 2025-05-06
AI Technical Summary
The prior art is difficult to achieve high accuracy in aerodynamic thermal prediction of deformed structure aircraft while reducing the calculation amount, resulting in large calculation resources and extended design cycles.
The client provides raw data to the computing server, the server builds a training sample database and trains a conditional diffusion model, and uses this model to predict and analyze the aircraft to be processed to generate aerodynamic thermal data on the object surface of the aircraft.
It realizes the rapid prediction of aerodynamic thermal data on the aircraft surface through inflow conditions and appearance conditions, which reduces the calculation amount, reduces the calculation time, and provides high-precision aerodynamic thermal prediction results.
Smart Images
Figure CN119940184A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of computational fluid dynamics analysis, and in particular to a method for rapid aerodynamic thermal analysis of a deformable structure aircraft surface. Background Art
[0002] Traditional aerodynamic thermal prediction methods include flight tests, wind tunnel tests, engineering algorithms, and computational fluid dynamics simulation (CFD). Although flight tests and wind tunnel tests can accurately predict aircraft aerodynamic thermal data, they require the manufacture of physical models or the application of wind tunnel test facilities. The cost of experimental analysis is very high, making it difficult to use them as conventional prediction and analysis methods.
[0003] Although numerical simulation methods can achieve high-precision aero-thermal predictions for complex-shaped aircraft, their high computational complexity, difficulty in convergence, and high requirements on mesh quality still cannot meet the needs of rapid and accurate prediction of aero-thermal in engineering design.
[0004] After cost compromise, the commonly used processing method is to approximate the real model based on test or numerical simulation data through regression analysis to obtain analytical expressions of aircraft aerodynamic parameters. However, these methods require high-quality CFD analysis and calculation processes. Each round of calculation process requires a large amount of computing resources and has high requirements for computer personnel. Especially for colleges and universities and scientific research institutes that mainly study the aerospace field, the occupation of computing resources is also very serious. There are often multiple project teams queuing up to wait for the allocation of computing resources, which significantly prolongs the design cycle of the aircraft.
[0005] Especially in the aerodynamic research of deformable aircraft, each deformed form of the aircraft requires a detailed aerodynamic analysis, and the amount of calculation and the time taken by the computing resources increase exponentially. Therefore, how to achieve high-precision aerodynamic thermal prediction analysis while reducing the amount of calculation has become a topic that needs further research and optimization. Summary of the invention
[0006] An embodiment of the present invention provides a rapid aerodynamic-thermal analysis method for a deformable structure aircraft surface, which can predict the aerodynamic-thermal data of the aircraft surface through the incoming flow conditions and shape conditions, reduce the amount of calculation and provide aerodynamic-thermal prediction results with higher accuracy.
[0007] To achieve the above object, the embodiments of the present invention adopt the following technical solutions:
[0008] A method for rapid aerodynamic thermal analysis of a deformable structure aircraft surface, wherein a client provides raw data to a computing server, and the method for rapid aerodynamic thermal analysis executed on the computing server comprises:
[0009] S1. Accepting raw data uploaded by a client, wherein the raw data includes aerodynamic thermal data of surface grid points of an aircraft sample;
[0010] S2, build a training sample database;
[0011] S3, training a conditional diffusion model using the training sample database;
[0012] S4, using the trained conditional diffusion model to perform prediction analysis on the aircraft to be processed, wherein the object of the prediction analysis includes the aerodynamic heat of the aircraft surface corresponding to the shape condition to be predicted and the incoming flow condition provided by the client;
[0013] S5. Send the prediction result to the client.
[0014] The aerodynamic thermal rapid prediction method for the deformable structure aircraft surface provided by the embodiment of the present invention uses CFD data, incoming flow conditions and shape conditions provided by the client to perform aerodynamic thermal prediction modeling analysis on the server side, first obtains the CFD original data, incoming flow conditions and shape conditions, builds a training sample database, trains the conditional diffusion model, and predicts the corresponding aircraft surface aerodynamic thermal data based on the incoming flow parameters and shape parameters to be predicted provided by the client, and finally returns the aircraft surface aerodynamic thermal data to the client. The aircraft surface aerodynamic thermal data can be predicted by the incoming flow conditions and shape conditions, reducing the amount of calculation and providing aerodynamic thermal prediction results with higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0016] Figure 1 A schematic diagram of a hardware environment provided for an embodiment of the present invention;
[0017] Figure 2 A flow chart of a method provided by an embodiment of the present invention;
[0018] Figure 3 A schematic diagram of the appearance of an aircraft provided in an embodiment of the present invention;
[0019] Figure 4 A schematic diagram of the mesh division of the aircraft surface provided by an embodiment of the present invention;
[0020] Figure 5 A schematic diagram of the conditional diffusion model structure provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0021] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present invention will be described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention. It can be understood by those skilled in the art that, unless specifically stated, the singular forms "one", "one", "said" and "the" used herein may also include plural forms. It should be further understood that the term "including" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used here may include wireless connection or coupling. The term "and / or" used herein includes any unit and all combinations of one or more associated listed items. It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as generally understood by those of ordinary skill in the art to which the present invention belongs. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.
[0022] The embodiment of the present invention provides a method for rapid prediction of aerodynamic heat of a deformable structure aircraft surface, the method is applied to a Figure 1 The computing resource allocation system composed of a client and a server as shown, wherein the client provides the server with raw data and aircraft shape conditions and incoming flow conditions, and the server constructs these data into a training sample database, and uses the sample database to train the conditional diffusion model, and then the incoming flow conditions and shape conditions to be predicted are input from the client, and the server sends the conditions to the trained conditional diffusion model to calculate the aerodynamic thermal data under the corresponding conditions, and the server feeds back the aerodynamic thermal prediction result information to the client. The aerodynamic thermal rapid prediction method, such as Figure 2 As shown, including:
[0023] S1. Accepting raw data uploaded by a client, wherein the raw data includes aerodynamic thermal data of surface grid points of an aircraft sample;
[0024] S2, build a training sample database;
[0025] S3, training a conditional diffusion model using the training sample database;
[0026] S4, using the trained conditional diffusion model to perform prediction analysis on the aircraft to be processed, wherein the object of the prediction analysis includes the aerodynamic heat of the aircraft surface corresponding to the shape condition to be predicted and the incoming flow condition provided by the client;
[0027] S5. Send the prediction result to the client.
[0028] Among them, the original data uploaded by the client also includes the shape conditions and incoming flow conditions of the aircraft sample, wherein the shape conditions and incoming flow conditions of the aircraft sample are used to perform simulation calculations to obtain the aerodynamic thermal data of the surface grid points of the aircraft sample in the original data; the data set in the training sample database includes: the incoming flow conditions, shape conditions and corresponding aerodynamic thermal data of the surface grid points of the aircraft sample, and the aerodynamic thermal data of the surface grid points of the aircraft sample include: the heat flux values corresponding to each surface grid node of the aircraft sample. For example: the shape conditions of the aircraft include: the spatial three-dimensional coordinates of the grid points on the surface of the aircraft and the wing extension and deflection angle; the incoming flow conditions of the aircraft include: the altitude, speed and angle of attack of the aircraft; the shape structure of the aircraft such as Figure 3 The aircraft shape parameters include the three-dimensional coordinates of the grid points and the wing extension and deflection angle. Figure 4 As shown. The specific original data of the aerodynamic thermal numerical simulation of the aircraft refers to: the aerodynamic thermal data of the surface of the corresponding object surface grid point under a certain working condition and a certain deflection angle of the aircraft wing. The original data is constructed as the model input, that is, the original data is expressed as: (α, H, V, AOA, x, y, z, q), α is the extension and deflection angle of the aircraft wing, H represents the height, V represents the speed, AOA represents the angle of attack, and x, y, z represent the three-dimensional coordinates in space. Among them, the deflection angle and three-dimensional spatial coordinates are the shape conditions, the height, speed and angle of attack are the incoming flow conditions, and q is the aerodynamic thermal data of the object surface grid point.
[0029] In this embodiment, the process of training the conditional diffusion model using the sample database includes:
[0030] The model input is used as the characteristic condition of the conditional diffusion model, and the heat flow value (aerodynamic thermal data) corresponding to the grid node of the aircraft sample is used as the prediction target of the conditional diffusion model. For example: the future flow conditions and shape conditions are used as the characteristic conditions of the conditional diffusion model, and the heat flow values corresponding to the grid nodes of the aircraft sample are used as the prediction target of the model to obtain data for model training. In the training stage, the future flow parameters and shape parameters are used as the conditional features of the conditional diffusion model, time embedding (artificially set) is used as the time step feature of the conditional diffusion model, and the grid point aerodynamic thermal data is used as the model prediction target (i.e., supervisory information). For example: in the training stage, the future flow conditions and shape conditions are used as the conditional features of the conditional diffusion model, expressed as: (α, H, V, AOA, x, y, z), and time embedding (time embedding) is used as the time step feature of the conditional diffusion model, and the aerodynamic thermal data q of the surface grid point is used as the model prediction target (also called supervisory information).
[0031] The conditional diffusion model includes two opposite processes, namely the forward diffusion process and the reverse denoising process. In the forward diffusion process, noise is continuously added to the aerodynamic thermal data of the aircraft surface. This process uses the reparameterization technique, where the data x after adding noise for the tth time (t has a total of 1000 steps) t It can be deduced by the recursive formula We get, where α t is a hyperparameter used to control the diffusion (noise) rate; ε t is a Gaussian noise until the aerodynamic thermal data of the object surface becomes pure Gaussian noise after multiple noise additions.
[0032] In the reverse denoising process, the added noise is continuously removed for the pure Gaussian noise obtained in the forward diffusion process, so as to restore it to the aerodynamic thermal data of the aircraft surface; this process can be performed by We get, where the variance is simply related to the hyperparameter α t The relevant constant (actually set to 0), ε θ To predict the noise, take the mean μ θ (x t ,t) as the predicted value of aerodynamic thermal data.
[0033] The conditional diffusion model adds characteristic conditions on the basis of the diffusion model, controls the denoising gradient of the model, and generates aerodynamic thermal data under corresponding characteristic conditions. for: where ε θ (y t ,t,x) is the conditional diffusion model denoising gradient, ε θ (yt ,t) is the unconditional diffusion model gradient, ω is used to control the weight between the two gradients. Based on this, the conditional diffusion model can be used to obtain the aerodynamic thermal data under the corresponding conditions and construct an aerodynamic thermal regression prediction model.
[0034] Furthermore, a fully connected neural network is used as a tool for predicting noise in the conditional diffusion model, wherein the loss function used in the training phase is: where y i represents the real noise, represents the prediction noise, and n is the number of samples.
[0035] In this embodiment, noise prediction is implemented based on a fully connected neural network to predict the noise added in each step of the forward process, including feature alignment, diffusion, dimensionality reduction, etc. The structure of the conditional diffusion model noise prediction includes: feature alignment module, diffusion module, dimensionality reduction module (all three modules use a fully connected neural network).
[0036] In the preferred scheme of this embodiment, the feature alignment module is composed of two layers of fully connected layers, the diffusion module is composed of three stacked diffusion modules, and the dimensionality reduction module is composed of three layers of fully connected layers. The structure of the input feature alignment module is: (7, 64, 1), 512. Among them, 7 represents the 7-dimensional feature conditions including the incoming flow parameters and the shape parameters (α, H, V, AOA, x, y, z), among which, compared with the model input, the part feature lacks one-dimensional aerodynamic thermal data q, q as the prediction target. 64 represents the dimension of time embedding (artificially set), 1 represents the one-dimensional object surface grid point aerodynamic thermal data q corresponding to the feature condition, and the feature alignment module upgrades the three different dimensional input features to 512 dimensions through a fully connected neural network;
[0037] The structure of the diffusion model denoising module is: 512, 512, feature fusion layer, 512, 512. This module linearly transforms the aligned features through two 512-dimensional fully connected layers, then fuses the features through the feature fusion layer, and then linearly transforms through two 512-dimensional fully connected layers. The diffusion module can be stacked multiple times.
[0038] The structure of the dimension reduction module is: 256, 64, 1. This module reduces the output features of the aforementioned diffusion module (several) to obtain predicted noise. The target dimension of the dimension reduction is consistent with the predicted aerodynamic thermal dimension, both of which are 1 dimension. The inverse denoising process is used The noise can be gradually removed from the Gaussian noise to generate aerodynamic thermal data of the object surface grid points under corresponding characteristic conditions.
[0039] For example, the structural diagram of the conditional diffusion model designed in this embodiment is as follows: Figure 5As shown in Figure 2, the aircraft shape parameters and incoming flow parameters are used as the characteristic conditions C of the diffusion model, and the aerodynamic thermal data q of the aircraft surface grid points are used as the prediction target to obtain the data matrix for model training.
[0040] The conditional diffusion model includes two opposite processes, namely forward diffusion and reverse denoising. Forward diffusion continuously adds noise to the original aerodynamic thermal data of the aircraft surface, and after several steps of denoising, it becomes pure Gaussian noise. Reverse denoising continuously removes the added noise from the data that has been denoised into a Gaussian distribution, restoring it to the original aerodynamic thermal data. The model obtains the distribution characteristics of the aerodynamic thermal data implicit in the denoising / denoising process through training. The noise prediction module is based on a fully connected neural network and includes: feature alignment module, diffusion module, and dimensionality reduction module. The noise prediction module is used to predict the noise added in each step of the forward process.
[0041] The input of the conditional diffusion model is: the predicted value y at step t t , time step t, seven-dimensional feature condition C. Since the dimensions of the three inputs are different, they need to be aligned through the fully connected layer; the diffusion module is used to enhance the ability to capture time steps and feature conditions; the dimensionality reduction module is used to reduce the dimensionality of the intermediate features and predict the one-dimensional noise ε t , and then the diffusion model is used to denoise the predicted value y at step t-1 t-1 , and then the next iteration is performed until y0 is generated, that is, the model predicts the aerodynamic thermal data according to the characteristic conditions.
[0042] In the preferred embodiment, the feature alignment module structure has two fully connected layers, namely (7, 64, 1), 512, where 7 represents the 7-dimensional feature conditions including the incoming flow parameters and the shape parameters (α, H, V, AOA, x, y, z), 64 represents the dimension of time embedding (artificially set), and 1 represents the aerodynamic thermal data q of the surface grid point corresponding to the feature condition. The input feature alignment module upgrades the input features of three different dimensions to 512 dimensions through a fully connected neural network; the diffusion module structure is 512, 512, feature fusion layer, 512, 512. The module linearly transforms the aligned features through two 512-dimensional fully connected layers, then fuses the features through a feature fusion layer, and then linearly transforms the features through two 512-dimensional fully connected layers. The diffusion module can be stacked multiple times.
[0043] The structure of the dimensionality reduction module is: 256, 64, 1. This module reduces the dimension of the final output features of the aforementioned diffusion module (several). The target dimension of the dimension reduction is consistent with the predicted aerodynamic thermal dimension, both of which are 1 dimension. The noise can be gradually removed from the Gaussian noise by using the inverse denoising process to generate the aerodynamic thermal data of the surface grid points under the corresponding characteristic conditions.
[0044] In the training phase: each round of iteration randomly selects a step size t to add noise to obtain the noisy data y t , the characteristic condition C and the noise step length t-1 are used as the input of the model to obtain the t-th step prediction noise ε t , train the model to approximate the real noise ε. The training loss function is: where y i represents the real noise, Represents prediction noise. In the prediction stage, the model gradually removes the prediction noise through Gaussian noise to achieve the prediction of aerodynamic thermal data.
[0045] The method for rapid aerodynamic-thermal prediction of the surface of an aircraft with a deformable structure provided in an embodiment of the present invention uses CFD data, incoming flow conditions, and shape conditions provided by a client to perform aerodynamic-thermal prediction modeling and analysis on the server side. The original CFD data, incoming flow conditions, and shape conditions are first obtained, a sample database is constructed, and a conditional diffusion model is trained. The trained conditional diffusion model is used to predict the corresponding aerodynamic-thermal data of the aircraft surface based on the incoming flow conditions and shape conditions to be predicted provided by the client, and finally the aerodynamic-thermal data of the aircraft surface under these conditions are returned to the client.
[0046] The main advantages of this embodiment are: fast construction of training sample database; short time period for training conditional diffusion model through sample database, easy to implement; aerodynamic thermal data prediction stage reduces a lot of calculation amount relative to simulation calculation, greatly reduces calculation time, can efficiently and accurately obtain aerodynamic thermal data of aircraft surface corresponding to shape conditions and incoming flow conditions; reduces dependence on simulation data. For example, to obtain aerodynamic thermal data of aircraft surface under certain shape conditions and incoming flow conditions, numerical simulation method requires 36 hours, while the aerodynamic thermal data prediction using the trained model of the present invention does not exceed two minutes.
[0047] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The above is only a specific implementation method of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or replacements that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A method for rapid aerodynamic thermal analysis of a deformable structure aircraft surface, characterized in that: The client provides raw data to the computing server, and the aerodynamic thermal rapid analysis method executed on the computing server includes: S1. Accepting raw data uploaded by a client, wherein the raw data includes aerodynamic thermal data of surface grid points of an aircraft sample; S2, build a training sample database; S3, training a conditional diffusion model using the training sample database; S4, using the trained conditional diffusion model to perform prediction analysis on the aircraft to be processed, wherein the object of the prediction analysis includes the aerodynamic heat of the aircraft surface corresponding to the shape condition to be predicted and the incoming flow condition provided by the client; S5. Send the prediction result to the client.
2. The method according to claim 1, characterized in that: The original data uploaded by the client also includes the shape conditions and incoming flow conditions of the aircraft sample, wherein the shape conditions and incoming flow conditions of the aircraft sample are used to obtain the aerodynamic thermal data of the object surface grid points of the aircraft sample in the original data; The data set in the training sample database includes: the incoming flow conditions, shape conditions and aerodynamic thermal data of the corresponding surface grid points of the aircraft samples, and the aerodynamic thermal data of the surface grid points of the aircraft samples include: the heat flux values corresponding to each surface grid node of the aircraft samples.
3. The method according to claim 2, characterized in that The shape conditions of the aircraft include: the spatial three-dimensional coordinates of the grid points on the surface of the aircraft and the wing extension and deflection angle; the incoming flow conditions of the aircraft include: the altitude, speed and angle of attack of the aircraft; The raw data is expressed as: (α, H, V, AOA, x, y, z, q), where α is the aircraft wing extension and deflection angle, H represents the height, V represents the speed, AOA represents the angle of attack, and x, y, z represent the three-dimensional coordinates in space. The deflection angle and three-dimensional spatial coordinates are shape conditions, the height, speed and angle of attack are the incoming flow conditions, and q is the aerodynamic thermal data of the surface grid point.
4. The method according to claim 3, characterized in that In S3, this includes: During the training phase, future flow conditions and shape conditions are used as conditional features of the conditional diffusion model, expressed as: (α, H, V, AOA, x, y, z), and time embedding is used as the time step feature of the conditional diffusion model, and the aerodynamic thermal data q of the surface grid points is used as the model prediction target.
5. The method according to claim 1 or 4, characterized in that: The conditional diffusion model includes two opposite processes, namely, a forward diffusion process and a reverse denoising process; Wherein, in the forward diffusion process, noise is continuously added to the aerodynamic thermal data of the object surface of the aircraft until the aerodynamic thermal data of the object surface becomes pure Gaussian noise after multiple noise additions, and the pure Gaussian noise is in the form of Gaussian distributed data; In the reverse denoising process, the added noise is continuously removed from the pure Gaussian noise obtained in the forward diffusion process to restore it to the aerodynamic thermal data of the aircraft surface.
6. The method according to claim 5, characterized in that Also includes: The conditional diffusion model adds characteristic conditions on the basis of the diffusion model, controls the denoising gradient of the model, and generates aerodynamic thermal data under corresponding characteristic conditions. for: where ε θ (y t ,t,x) is the denoising gradient of the conditional diffusion model, ε θ (y t ,t) is the unconditional diffusion model gradient, ω is used to control the weight between the two gradients. Based on this, the conditional diffusion model can be used to obtain the aerodynamic thermal data under the corresponding conditions and construct an aerodynamic thermal regression prediction model.
7. The method according to claim 6, characterized in that Also includes: A fully connected neural network is used as a tool for predicting noise in the conditional diffusion model, wherein the loss function used in the training phase is: y i represents the true value of the i-th sample, represents the predicted value of the i-th sample, and n is the number of samples.
8. The method according to claim 6, characterized in that The input feature alignment module consists of two fully connected layers, the diffusion model denoising module consists of three stacked diffusion modules, and the intermediate feature dimensionality reduction module consists of three fully connected layers.
9. The method according to claim 8, characterized in that The structure of the input feature alignment module is: (7, 64, 1), 512; The structure of the diffusion model denoising module is: 512, 512, feature fusion layer, 512, 512; The structure of the intermediate feature dimensionality reduction module is: 256, 64, 1.