A method for aerodynamic thermal prediction based on support vector machine and boundary layer theory

Abstract

The present invention relates to an aerodynamic heat prediction method based on support vector machine and boundary layer theory. It is known from boundary layer theory that aerodynamic heat has strong locality and is a local quantity, and its distribution depends to a large extent on the boundary layer outer edge information. Based on the close relationship between boundary layer outer edge information and aerodynamic heat, this paper adopts a data-driven localized modeling approach and uses support vector machine to construct an aerodynamic heat prediction model from local boundary layer outer edge characteristics to local wall aerodynamic heat. The benefit of this method is that there is no need to draw a finer grid to solve the time-consuming NS equation. It is only necessary to directly solve the simple and time-saving Euler equation to obtain the boundary layer outer edge characteristics, and the aerodynamic heat distribution can be quickly obtained through the trained model. Compared with traditional numerical calculation methods, it avoids the uncertainty of grid quality and improves efficiency by more than ten times while ensuring prediction accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of aerospace technology, and relates to an aerodynamic heat prediction method based on a support vector machine and boundary layer theory, and specifically relates to a prediction of aerodynamic heat distribution on the surface of a hypersonic aircraft. Background Art

[0002] During the flight of a hypersonic aircraft, gas compression and friction will cause high-temperature airflow on the aircraft surface. The heat of the high-temperature airflow is further transferred to the aircraft, heating or even burning the aircraft structure. For the safety of hypersonic aircraft, the aerodynamic heat distribution on the aircraft surface during flight must be considered when designing the aircraft, so aerodynamic heat prediction is required. However, the increasingly complex shape and tight design cycle of hypersonic aircraft make it difficult for existing methods to meet the needs of efficient and accurate aerodynamic heat prediction.

[0003] Traditional aero-thermal prediction methods can be mainly divided into two categories: experimental methods and computational methods. The experimental methods include flight tests and ground wind tunnel tests, while the computational methods are mainly divided into engineering algorithms and numerical simulation methods. Although flight tests can accurately predict the aero-thermal performance of aircraft in flight, they are difficult to use as a routine research method due to their high cost. Ground wind tunnel tests face the problem of consistency between the sky and the earth, and it is also difficult to conduct large-scale aero-thermal predictions due to cost constraints (Peng Zhiyu, Shi Yilei, Gong Hongming, et al. Hypersonic aero-thermal prediction technology and development trends [J]. Acta Aeronautica Sinica, 2015, 36(01): 325-345; Luo Changtong, Hu Zongmin, Liu Yunfeng, et al. Research progress on the correlation between the sky and the earth of hypersonic wind tunnel aerodynamic / thermal test data [J]. Experimental Fluid Mechanics, 2020, 34(03): 78-89.). Engineering methods with experimental and theoretical foundations can efficiently predict aerodynamic heat, but because they contain many assumptions, the accuracy of aerodynamic heat prediction for complex shapes and flows is low (Lü Lili, Zhang Weiwei, Ye Zhengyin. Calculation of surface heat flux of hypersonic reentry bodies [J]. Journal of Applied Mechanics, 2006, 23(2): 259-262.). Although numerical simulation methods can achieve high-precision aerodynamic heat prediction for complex-shaped aircraft, they have high requirements on mesh quality, slow convergence speed, and large amount of calculation, and still cannot meet the needs of rapid and accurate prediction of aerodynamic heat in engineering design (Wang Ziyi, Zhang Weiwei, Liu Lei. A review of thermoaeroelastic simulation calculation methods for hypersonic aircraft [J]. Gas Physics, 2020, 5(06): 1-15; Yu Chengzhang, Liu Weihua. Research progress in aerodynamic heat prediction technology for hypersonic aircraft [J]. Aeronautical Science and Technology, 2021, 32(02): 14-21.).

[0004] Patent CN113901594B discloses an intelligent prediction method for aerodynamic heat of the surface of a hypersonic aircraft using autoencoder technology, but this method requires the use of a large amount of aerodynamic heat data to build an aerodynamic heat database. Patent CN113821877A discloses an aerodynamic heat numerical simulation method using the regional decomposition calculation idea, and patent CN111339672A discloses an aerodynamic heat simulation analysis method for obtaining the inlet leading edge shock wave temperature field by statistically calculating the thermodynamic state parameters of simulated molecules. Both patents use numerical simulation methods to calculate aerodynamic heat. Patent CN106872195B discloses a method for correlation analysis of aerodynamic heat data between different typical positions of a high-speed aircraft that can reduce the number of test tests, but this method cannot achieve aerodynamic heat prediction of aircraft under different incoming flow conditions. In summary, there is currently no public patent or literature that uses the support vector machine algorithm and boundary layer theory to achieve aerodynamic heat prediction. Summary of the invention

[0005] Technical issues to be solved

[0006] In order to avoid the shortcomings of the prior art, the present invention proposes an aerodynamic heat prediction method based on support vector machine and boundary layer theory, which is aimed at the dilemma that the demand for rapid and accurate prediction of aerodynamic heat in engineering design is difficult to meet.

[0007] Technical Solution

[0008] An aerodynamic heat prediction method based on support vector machine and boundary layer theory is characterized by the following steps:

[0009] Step 1: Solve the Euler equation of the aircraft to obtain the outer edge information of the boundary layer of the aircraft under a certain incoming flow state; at the same time, use the RANS method to calculate the aerodynamic heat distribution sample of the aircraft under the same incoming flow state;

[0010] Step 2: Construct a candidate boundary layer outer edge feature library:

[0011] The first method is to use the density of the incoming flow condition parameter (ρ ∞ ), pressure (P ∞ ), temperature (T ∞ ), speed (U ∞ ), the information density at the outer edge of the boundary layer (ρ e ), pressure (P e ), temperature (T e ), speed (U e ), viscosity coefficient (μ e ), wall parameter temperature (T w ) is directly dimensionless;

[0012] The second method is to calculate the difference between the incoming flow parameters, boundary layer outer edge parameters and wall parameters, and then convert them into dimensionless values.

[0013] The third type: structural features based on the appearance characteristics of the aircraft

[0014] Step 3: Determine the input boundary layer outer edge feature combination from the candidate boundary layer outer edge feature library by using a feature selection method;

[0015] Step 4: Using the boundary layer outer edge feature combination as input and the aerodynamic heat distribution sample as output, a model training data set is constructed;

[0016] Step 5: Use the training data set to train the support vector machine to achieve the mapping of boundary layer outer edge characteristics and wall aerodynamic heat, complete the construction of the aerodynamic heat prediction model, and output the model training accuracy

[0017] Step 6: Solve the Euler equation of the aircraft under a certain incoming flow state, which is the same as the incoming flow state of the aircraft's aerodynamic thermal distribution that needs to be predicted; construct a boundary layer outer edge feature combination, which is the same as the final boundary layer outer edge combination of the model; input the boundary layer outer edge feature combination into the established aerodynamic thermal model to obtain the aerodynamic thermal distribution of the aircraft that needs to be predicted.

[0018] The first dimensionless method of constructing the candidate boundary layer outer edge feature library is to calculate the ratio of the corresponding physical quantity; or the classical dimensionless parameter Mach number Reynolds number Dimensionless.

[0019] The second type of constructing the candidate boundary layer outer edge feature library: feature (T e -T w ) / T ∞ .

[0020] The third type of constructing the candidate boundary layer outer edge feature library: local angle of attack α s =tan -1 (v e / u e ),u e is the velocity in the x direction at the outer edge of the boundary layer, v e is the velocity in the y direction at the outer edge of the boundary layer.

[0021] The specific steps of step 3 are:

[0022] Step 3.1: Use the random forest method to perform importance analysis on the candidate features and sort them from most important to least important.

[0023] Step 3.2: Starting from the candidate boundary layer outer edge feature with the greatest importance, the number of features in the candidate boundary layer outer edge feature combination is increased one by one in descending order of importance, so as to obtain different candidate boundary layer outer edge feature combinations;

[0024] Step 3.3: Then, different candidate boundary layer outer edge feature combinations are respectively subjected to step 4 and step 5 until the model training accuracy is no longer improved or all candidate boundary layer outer edge feature combinations are traversed;

[0025] Step 3.4: Finally, the candidate boundary layer outer edge feature combination with the highest model training accuracy is selected as the final input boundary layer outer edge feature combination of the model.

[0026] In step 3.2: the first candidate boundary layer outer edge feature combination only includes the candidate boundary layer outer edge feature with the greatest importance, the second candidate boundary layer outer edge feature combination includes the candidate boundary layer outer edge features with the greatest importance and the second greatest importance, the third candidate boundary layer outer edge feature combination includes the candidate boundary layer outer edge features with the greatest importance, the second greatest importance and the third greatest importance, and so on to obtain different candidate boundary layer outer edge feature combinations.

[0027] Beneficial Effects

[0028] The present invention proposes an aerodynamic heat prediction method based on support vector machine and boundary layer theory. It is known from boundary layer theory that aerodynamic heat has strong locality and is a local quantity, and its distribution depends to a large extent on the boundary layer outer edge information. Based on the close relationship between boundary layer outer edge information and aerodynamic heat, this paper adopts a data-driven localized modeling approach and uses support vector machine to construct an aerodynamic heat prediction model from local boundary layer outer edge characteristics to local wall aerodynamic heat. The benefit of this method is that there is no need to draw a finer grid to solve the time-consuming NS equation. It is only necessary to directly solve the simple and time-saving Euler equation to obtain the boundary layer outer edge characteristics, and the aerodynamic heat distribution can be quickly obtained through the trained model. Compared with traditional numerical calculation methods, it avoids the uncertainty of grid quality and improves efficiency by more than ten times while ensuring prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Schematic diagram of the process of building the aerodynamic heat prediction model in the method of the present invention

[0030] Figure 2 Schematic diagram of feature selection process in the method of the present invention DETAILED DESCRIPTION

[0031] The present invention will now be further described with reference to the embodiments and the accompanying drawings:

[0032] The technical solution adopted by the present invention to solve the technical problem comprises the following steps:

[0033] Step 1: Sample data acquisition. Solve the Euler equation of the aircraft to obtain the outer edge information of the boundary layer of the aircraft under a certain incoming flow state, and use the RANS method to calculate the aerodynamic heat distribution samples of the aircraft under the same incoming flow state.

[0034] Step 2: Construct a candidate boundary layer outer edge feature library. There are three main methods for constructing candidate features. The first method is to use the density of the incoming flow condition parameter (ρ ∞ ), pressure (P ∞ ), temperature (T ∞ ), speed (U ∞ ), the information density at the outer edge of the boundary layer (ρ e ), pressure (P e ), temperature (T e ), speed (U e ), viscosity coefficient (μ e ), wall parameter temperature (T w ) is directly dimensionless. The dimensionless method is to find the ratio of the corresponding physical quantity (such as ρ e / ρ ∞ ), following the classical dimensionless parameter Mach number Reynolds number The second method is to calculate the difference between the incoming flow parameters, boundary layer outer edge parameters and wall parameters, and then make them dimensionless, such as the characteristic (T e -T w ) / T ∞ The third type: based on the structural characteristics of the aircraft shape, such as the local angle of attack α s =tan -1 (v e / u e ),u e is the velocity in the x direction at the outer edge of the boundary layer, v e is the velocity in the y direction at the outer edge of the boundary layer.

[0035] Step 3: Determine the input boundary layer outer edge feature combination. Determine the input boundary layer outer edge feature combination from the candidate boundary layer outer edge feature library through a feature selection method.

[0036] Step 4: Construct a model training data set. The model training data set is constructed with the boundary layer outer edge feature combination as input and the aerodynamic heat distribution sample as output.

[0037] Step 5: Aerothermal prediction model construction. Use the training data set to train the support vector machine to achieve the mapping of boundary layer outer edge characteristics and wall aerodynamic heat, complete the construction of the aerothermal prediction model, and output the model training accuracy.

[0038] Step 6: Aero-thermal prediction of aircraft. Solve the Euler equation of the aircraft under a certain incoming flow state, which is the same as the incoming flow state of the aircraft aero-thermal distribution to be predicted. Construct a boundary layer outer edge feature combination, which is the same as the final boundary layer outer edge combination of the model. Input the boundary layer outer edge feature combination into the established aero-thermal model to obtain the aero-thermal distribution of the aircraft to be predicted.

[0039] Wherein, the specific steps of step 3 of the present invention are:

[0040] Step 3.1: Use the random forest method to perform importance analysis on the candidate features and sort them from most important to least important.

[0041] Step 3.2: Starting from the candidate boundary layer outer edge feature with the greatest importance, the number of features in the candidate boundary layer outer edge feature combination is increased one by one in descending order of importance to obtain different candidate boundary layer outer edge feature combinations.

[0042] The first candidate boundary layer outer edge feature combination only includes the candidate boundary layer outer edge feature with the greatest importance, the second candidate boundary layer outer edge feature combination includes the candidate boundary layer outer edge features with the greatest importance and the second greatest importance, the third candidate boundary layer outer edge feature combination includes the candidate boundary layer outer edge features with the greatest importance, the second greatest importance, and the third greatest importance, and so on to obtain different candidate boundary layer outer edge feature combinations.

[0043] Step 3.3: Then, different candidate boundary layer outer edge feature combinations are respectively subjected to step 4 and step 5 until the model training accuracy is no longer improved or all candidate boundary layer outer edge feature combinations are traversed.

[0044] Step 3.4: Finally, the candidate boundary layer outer edge feature combination with the highest model training accuracy is selected as the final input boundary layer outer edge feature combination of the model. Specific embodiment:

[0046] like Figure 1 and Figure 2 As shown, an aerodynamic heat prediction method based on support vector machine and boundary layer theory includes the following steps:

[0047] Step 1: Solve the Euler equation of the double ellipsoid shape to obtain the outer edge information of the centerline boundary layer on the upper surface of the double ellipsoid at 0°, 10°, 20° and 25° angles of attack. Use the RANS method to calculate the aerodynamic heat distribution samples of the centerline of the upper surface of the double ellipsoid under the same incoming flow state.

[0048] Step 2: Construct a candidate boundary layer outer edge feature library. The results are shown in Table 1. In Table 1, s is the streamline length, that is, the distance from any point on the center line of the double ellipsoid surface along the wall to the leading edge point, α s =tan-1 (v e / u e ) is the local angle of attack, u is the speed in the x direction, v is the speed in the y direction, and the speed ρ is air density, P is air pressure, T is temperature, μ is viscosity, γ is specific heat ratio, R is air constant, Ma is Mach number, subscript e represents boundary layer outer edge variable, subscript ∞ represents incoming flow variable, and subscript w represents wall variable.

[0049] Step 3: Determine the boundary layer outer edge feature combination as [f2, f3, f1, f6, f5] from the candidate boundary layer outer edge feature library through the feature selection method.

[0050] Step 4: Extract the boundary layer outer edge features and aerodynamic thermal data of 65 points on the center line of the upper surface of the double ellipsoid at each angle of attack, and construct a model training data set with the boundary layer outer edge features as input and the aerodynamic thermal distribution samples as output.

[0051] Step 5: Use the training data set to train the support vector machine to achieve the mapping of the boundary layer outer edge characteristics and the wall aerodynamic heat distribution, complete the construction of the aerodynamic heat prediction model, and output the model training accuracy.

[0052] Step 6: Solve the Euler equation for the center line of the upper surface of the double ellipsoid at 5°, 15° and 30° angles of attack to obtain the boundary layer outer edge feature combination, input the boundary layer outer edge feature combination into the established aerodynamic thermal model, and the aerodynamic thermal distribution of the center line of the upper surface of the double ellipsoid at 5°, 15° and 30° angles of attack can be predicted.

[0053] Wherein, the specific steps of step 3 of the present invention are:

[0054] Step 3.1: Use the random forest method to analyze the importance of candidate features and sort them from most important to least important. The final feature ranking is f2>f3>f1>f6>f5>f8>f9>f4>f7.

[0055] Step 3.2: Starting from the candidate boundary layer outer edge feature f2 with the greatest importance, the number of features in the candidate boundary layer outer edge feature combination is increased one by one in descending order of importance to obtain different candidate boundary layer outer edge feature combinations.

[0056] Step 3.3: Then, different candidate boundary layer outer edge feature combinations are respectively subjected to step 4 and step 5 until the model training accuracy is no longer improved or all candidate boundary layer outer edge feature combinations are traversed. After selecting the candidate boundary layer outer edge feature combination [f2, f3, f1, f6, f5], the model training accuracy is no longer improved.

[0057] Step 3.4: Finally, select the boundary layer outer edge feature combination [f2, f3, f1, f6, f5] with the highest model training accuracy as the final input boundary layer outer edge feature combination of the model.

[0058] Table 1 Candidate boundary layer outer edge features

[0059]

Claims

1. An aerodynamic heat prediction method based on support vector machine and boundary layer theory, characterized in that Here are the steps: Step 1: Solve the Euler equation of the aircraft to obtain the outer edge information of the boundary layer of the aircraft under a certain incoming flow state; at the same time, use the RANS method to calculate the aerodynamic heat distribution sample of the aircraft under the same incoming flow state; Step 2: Construct a candidate boundary layer outer edge feature library: The first method is to use the density of the incoming flow condition parameter (ρ ∞ ), pressure (P ∞ ), temperature (T ∞ ), speed (U ∞ ), the information density at the outer edge of the boundary layer (ρ e ), pressure (P e ), temperature (T e ), speed (U e ), viscosity coefficient (μ e ), wall parameter temperature (T w ) is directly dimensionless; The second method is to calculate the difference between the incoming flow parameters, boundary layer outer edge parameters and wall parameters, and then convert them into dimensionless values. The third type: structural features based on the appearance characteristics of the aircraft: Step 3: Determine the input boundary layer outer edge feature combination from the candidate boundary layer outer edge feature library by using a feature selection method; Step 4: Using the boundary layer outer edge feature combination as input and the aerodynamic heat distribution sample as output, a model training data set is constructed; Step 5: Use the training data set to train the support vector machine to achieve the mapping between the outer edge characteristics of the boundary layer and the wall aerodynamic heat, complete the construction of the aerodynamic heat prediction model, and output the model training accuracy; Step 6: Solve the Euler equation of the aircraft under a certain incoming flow state, which is the same as the incoming flow state of the aircraft's aerodynamic thermal distribution that needs to be predicted; construct a boundary layer outer edge feature combination, which is the same as the final boundary layer outer edge combination of the model; input the boundary layer outer edge feature combination into the established aerodynamic thermal model to obtain the aerodynamic thermal distribution of the aircraft that needs to be predicted.

2. The aerodynamic heat prediction method based on support vector machine and boundary layer theory according to claim 1 is characterized in that: The first dimensionless method of constructing the candidate boundary layer outer edge feature library is to calculate the ratio of the corresponding physical quantities; or the classical dimensionless parameter Mach number Reynolds number Dimensionless.

3. The aerodynamic heat prediction method based on support vector machine and boundary layer theory according to claim 1 is characterized in that: The second type of constructing the candidate boundary layer outer edge feature library: feature (T e -T w ) / T ∞ .

4. The aerodynamic heat prediction method based on support vector machine and boundary layer theory according to claim 1 is characterized in that: The third type of constructing the candidate boundary layer outer edge feature library: local angle of attack α s =tan -1 (v e / u e ),u e is the velocity in the x direction at the outer edge of the boundary layer, v e is the y-direction velocity at the outer edge of the boundary layer.

5. The aerodynamic heat prediction method based on support vector machine and boundary layer theory according to claim 1 is characterized in that: The specific steps of step 3 are: Step 3.1: Use the random forest method to analyze the importance of candidate features and sort them from most important to least important; Step 3.2: Starting from the candidate boundary layer outer edge feature with the greatest importance, the number of features in the candidate boundary layer outer edge feature combination is increased one by one in descending order of importance, so as to obtain different candidate boundary layer outer edge feature combinations; Step 3.3: Then, different candidate boundary layer outer edge feature combinations are respectively subjected to step 4 and step 5 until the model training accuracy is no longer improved or all candidate boundary layer outer edge feature combinations are traversed; Step 3.4: Finally, the candidate boundary layer outer edge feature combination with the highest model training accuracy is selected as the final input boundary layer outer edge feature combination of the model.

6. The aerodynamic heat prediction method based on support vector machine and boundary layer theory according to claim 5 is characterized in that: In step 3.2: the first candidate boundary layer outer edge feature combination only includes the candidate boundary layer outer edge feature with the greatest importance, the second candidate boundary layer outer edge feature combination includes the candidate boundary layer outer edge features with the greatest importance and the second greatest importance, the third candidate boundary layer outer edge feature combination includes the candidate boundary layer outer edge features with the greatest importance, the second greatest importance and the third greatest importance, and so on to obtain different candidate boundary layer outer edge feature combinations.

Images