Aerodynamic thermal environment proxy model construction method based on neural network
By constructing a neural network-based aerodynamic thermal environment proxy model, the problems of large computational load and long cycle in aerodynamic thermal environment simulation analysis in flight trajectory design are solved, enabling rapid and accurate prediction of surface heat flow of high-speed aircraft and improving the accuracy and applicability of the model.
Patent Information
- Application Number
- CN202511365273.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-23
- Publication Date
- 2026-01-23
AI Technical Summary
Existing technologies cannot take into account the aerodynamic and thermal environment of high-speed aircraft in real time during flight trajectory design. This results in a large amount of computation and a long cycle for the simulation analysis of the aerodynamic and thermal environment of complex three-dimensional aircraft, and it is impossible to quickly and accurately provide the surface heat flow.
A neural network-based aerodynamic thermal environment proxy model is constructed. By establishing a flight state sample space, aerodynamic thermal engineering calculations and data classification are performed. A fully connected neural network is trained to obtain the neural network model parameters. A neural network proxy model is then constructed to quickly provide the surface heat flow of typical parts of the aircraft.
It enables the rapid and accurate determination of aircraft surface heat flux in flight trajectory design, improving the model's accuracy and generalization performance, and is able to handle nonlinear processes of laminar-transition-turbulent flow state transition.
Smart Images
Figure CN121389424A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to a neural network-based aerodynamic thermal environment proxy model construction method, belonging to the field of simulation and design of high-speed aircraft aerodynamic heat and thermal protection system. BACKGROUND
[0002] When the aircraft flies at high speed in the atmosphere, it rubs against the surrounding air and compresses the air in front. Under the double effects of viscous dissipation and strong shock compression, a large amount of kinetic energy of the high-speed incoming flow is converted into internal energy, causing the temperature in the shock layer to rise sharply, which can reach more than 10,000 K, forming serious aerodynamic heating on the surface of the aircraft. The severe aerodynamic thermal environment forms a strong constraint on the flight corridor thermal boundary of the aircraft. When designing the flight trajectory, the thermal environment constraint is considered to ensure the reliability of the thermal protection system of the aircraft. However, the simulation and analysis of the three-dimensional complex aircraft aerodynamic thermal environment requires a large amount of calculation and a long period, which cannot be integrated into the flight trajectory design program in real time. Therefore, there is a demand for an aerodynamic thermal environment proxy model, which needs to quickly and accurately give the surface heat flux according to the flight state parameters. SUMMARY
[0003] The technical problem to be solved by the present application is: for the coupling design problem of high-speed aircraft thermal environment and flight trajectory, the present application provides a neural network-based aerodynamic thermal environment proxy model construction method, which can quickly and accurately give the surface heat flux of the typical parts of the aircraft according to the flight state parameters, and is used for flight trajectory optimization.
[0004] The technical scheme adopted by the present application is: a neural network-based aerodynamic thermal environment proxy model construction method, comprising:
[0005] A flight state sample space is established, so that the flight state sample space covers the flight profile parameter range of the aircraft to be calculated;
[0006] The heat flux and transition intermittency factor data of the state points of the selected typical parts of the aircraft are calculated;
[0007] According to the transition intermittency factor data, the flow state corresponding to the flight state sample points in the flight state sample space is determined;
[0008] According to the flow state corresponding to the flight state sample points in the flight state sample space, the heat flux data of the state points of the selected typical parts of the aircraft are classified to obtain a training data set;
[0009] The transition state sample space composed of the transition state sample points is encrypted to obtain the transition intermittency factor data of all state points in the encrypted transition state sample space;
[0010] construct a full connection neural network and train to obtain neural network model parameters;
[0011] According to the neural network model parameters, a neural network proxy model is constructed.
[0012] Further, in the flight state sample space, the flight altitude interval is not more than 2km, the Mach number interval is not more than 2, and the attack angle interval is not more than 2°. The flight altitude H i , the Mach number M j , and the attack angle α k are combined to obtain the flight state sample space [H i , M j , α k ], i, j, and k are positive integers.
[0013] Further, the calculation obtains the heat flow and transition intermittency factor data of the selected state points of the aircraft typical parts, including:
[0014] According to the aircraft shape to be calculated, the end radius, the leading edge radius, the leading edge sweep angle, and the centerline cone angle geometric parameters are determined as the shape input of the aerodynamic heat engineering calculation. The aerodynamic heat engineering calculation is performed on all state points in the flight state sample space to obtain the heat flow and transition intermittency factor data of all state points of the selected aircraft typical parts.
[0015] Further, the determination of the flow state corresponding to the flight state sample point in the flight state sample space includes:
[0016] The flight state sample point with a transition intermittency factor of 0 is a laminar flow state sample point, the flight state sample point with a transition intermittency factor of 1 is a turbulent flow state sample point, and the flight state sample point with a transition intermittency factor between 0 and 1 is a transition state sample point.
[0017] Further, the classification of the heat flow data of the selected state points of the aircraft typical parts according to the flow state corresponding to the flight state sample point in the flight state sample space obtains a training data set, including:
[0018] The data of the laminar flow state sample points are combined into a laminar flow state training data set, the data of the turbulent flow state sample points are combined into a turbulent flow state training data set, and the data of the transition state sample points are used as a transition state training data set.
[0019] Further, the transition state sample space composed of the transition state sample points is encrypted to obtain the transition intermittency factor data of all state points in the encrypted transition state sample space, including:
[0020] After the transition state sample space is encrypted, the flight height interval of the data in the transition state sample space is not more than 0.2 km, the Mach number interval is not more than 0.5, the attack angle interval is not more than 0.5°, and the transition state sample space state points after encryption are subjected to aerodynamic heat engineering calculation again to obtain the transition interval factor data of all state points in the encrypted transition state sample space.
[0021] Further, the full connection neural network is constructed and trained to obtain neural network model parameters, including:
[0022] A 4-layer full connection neural network is constructed, wherein the input layer includes 3 variables: height, Mach number and attack angle, the output layer is heat flow, and each of the two hidden layers in the middle includes 10 hidden variables; the flight state [H i , M j , alpha k ] in the laminar flow state sample space is input, and the heat flow is output; the full connection neural network is trained by using the back propagation algorithm for the laminar flow state training data set to obtain the laminar flow state heat flow neural network model parameters;
[0023] A 4-layer full connection neural network is constructed, wherein the input layer includes 3 variables: height, Mach number and attack angle, the output layer is heat flow, and each of the two hidden layers in the middle includes 10 hidden variables; the flight state [H i , M j , alpha k ] in the turbulent flow state sample space is input, and the heat flow is output; the full connection neural network is trained by using the back propagation algorithm for the turbulent flow state training data set to obtain the turbulent flow state heat flow neural network model parameters;
[0024] A 4-layer full connection neural network is constructed, wherein the input layer includes 3 variables: height, Mach number and attack angle, the output layer is the interval factor, and each of the two hidden layers in the middle includes 10 hidden variables; the flight state [H i , M j , alpha k ] in the transition state sample space is input, and the interval factor is output; the neural network is trained by using the back propagation algorithm for the transition interval factor training data set to obtain the interval factor neural network model parameters.
[0025] Further, the neural network proxy model is constructed according to the neural network model parameters, including:
[0026] For any given new state [height H new , Mach number M new , attack angle alpha new ] in the flight parameter range, the new state [H new , M new , alphanew As input, the laminar flow state heat flow neural network is used to calculate the laminar flow heat flow Q L , the turbulent flow state heat flow neural network is used to calculate the laminar flow heat flow Q T , the intermittent factor neural network is used to calculate the intermittent factor γ, and γ·Q L +(1-γ)·Q T is used as the neural network proxy model heat flow output of the aerodynamic heat environment.
[0027] Compared with the prior art, the present application has the following advantages:
[0028] (1) The aerodynamic method of the present application can obtain accurate aerodynamic heat environment values and laws based on a large number of aerodynamic heat engineering calculation results, and can integrate typical state numerical simulation results and test results to improve the model precision.
[0029] (2) The model of the present application has better generalization performance for flight states, and the model prediction accuracy is higher. The proxy model proposed in the present application has strong modeling ability for strong nonlinear processes such as laminar-turbulent flow state conversion. BRIEF DESCRIPTION OF DRAWINGS
[0030] Figure 1 is the flow chart of the present application. DETAILED DESCRIPTION
[0031] The present application is further described.
[0032] As Figure 1 shown, a neural network-based aerodynamic heat environment proxy model construction method comprises the following steps:
[0033] 1. Establish a flight altitude H i , Mach number M j , angle of attack α k sample space, so that the sample space covers the flight profile parameter range of the aircraft to be calculated, and the altitude interval is not more than 2km, the Mach number interval is not more than 2, and the angle of attack interval is not more than 2°. The altitude, Mach number and angle of attack sample points are cross combined to obtain a flight state sample space [H i , M j , α k ], i, j, k are all positive integers.
[0034] 2. According to the actual aircraft shape, determine the nose radius, leading edge radius, leading edge sweep angle, centerline cone angle geometric parameters as the shape input of aerodynamic heat engineering calculation, and perform aerodynamic heat engineering calculation for all state points in the flight state sample space determined in step 1 to obtain the heat flow and transition intermittent factor data of all state points of the selected aircraft typical parts.
[0035] 3. Determine the flow state corresponding to the flight state sample space in step 1, the flight state sample point with a transition intermittency factor of 0 is a laminar flow state sample point, the flight state sample point with a transition intermittency factor of 1 is a turbulent flow state sample point, and the flight state sample point with a transition intermittency factor between 0 and 1 is a transition state sample point;
[0036] 4. Classify the data calculated in step 2 according to the flow state determined in step 3, combine the data of the laminar flow state into a laminar flow state training data set, combine the data of the turbulent flow state into a turbulent flow state training data set, and take the data of the transition state as a transition state training data set.
[0037] 5. Encrypt the transition state sample space in step 3 so that the height interval is not more than 0.2 km, the Mach number interval is not more than 0.5, and the angle of attack interval is not more than 0.5°, and perform aerodynamic heat engineering calculation on the state points in the encrypted transition state sample space to obtain the transition intermittency factor data of all state points in the encrypted transition state sample space.
[0038] 6. Construct a 4-layer fully connected neural network, wherein the input layer includes 3 variables: height, Mach number, and angle of attack, the output layer is heat flow, and the two hidden layers each contain 10 hidden variables; take the flight state [H i , M j , α k ] in the laminar flow state sample space as input and the heat flow as output, train the fully connected neural network based on the laminar flow state training data set in step 4 using the back propagation algorithm, and obtain the laminar flow state heat flow neural network model parameters.
[0039] 7. Construct a 4-layer fully connected neural network, wherein the input layer includes 3 variables: height, Mach number, and angle of attack, the output layer is heat flow, and the two hidden layers each contain 10 hidden variables; take the flight state [H i , M j , α k ] in the turbulent flow state sample space as input and the heat flow as output, train the fully connected neural network based on the turbulent flow state training data set in step 4 using the back propagation algorithm, and obtain the turbulent flow state heat flow neural network model parameters.
[0040] 8. Construct a 4-layer fully connected neural network, wherein the input layer includes 3 variables: height, Mach number, and angle of attack, the output layer is intermittency factor, and the two hidden layers each contain 10 hidden variables; take the flight state [H i , M j , α k ] in the transition state sample space as input and the intermittency factor as output, train the neural network based on the transition intermittency factor training data set in step 4 using the back propagation algorithm, and obtain the intermittency factor neural network model parameters.
[0041] 9. Construct a neural network proxy model as follows: for any given new state [H new , M new , α new ] in the flight parameter range, take the new state as input, calculate the laminar heat flux Q L using the laminar state heat flux neural network obtained in step 6, calculate the laminar heat flux Q T using the turbulent state heat flux neural network constructed in step 7, calculate the intermittency factor γ using the intermittency factor neural network constructed in step 8, and take γ·Q L +(1-γ)·Q T as the aerodynamic heat environment proxy model heat flux output.
[0042] The method provided by the present application is not only suitable for constructing an aerodynamic heat environment proxy model of a high-speed aircraft, but also suitable for constructing an aerodynamic heat environment proxy model of other types of aircraft, and constructing other similar surface physical quantities. The above-mentioned embodiments are only used to explain the present application, and cannot be used as a limitation of the present application, so any similar implementation manner within the idea of the present application is within the protection scope of the present application.
[0043] The part not described in detail in the present application belongs to the known technology of those skilled in the art.
Claims
1. A method for constructing a neural network-based aerodynamic heating environment proxy model, characterized in that, The application relates to a method for constructing a neural network proxy model for a selected typical part of an aircraft, and belongs to the technical field of aircraft aerodynamic thermal engineering. The method comprises the following steps: establishing a flight state sample space, so that the flight state sample space covers the flight profile parameter range of an aircraft to be calculated; calculating thermal flow and transition intermittency factor data of state points of the selected typical part of the aircraft; determining flow states corresponding to flight state sample points in the flight state sample space according to the transition intermittency factor data; classifying thermal flow data of the state points of the selected typical part of the aircraft according to the flow states corresponding to the flight state sample points in the flight state sample space, so as to obtain a training data set; encrypting a transition state sample space formed by the transition state sample points, so as to obtain transition intermittency factor data of all state points in the encrypted transition state sample space; constructing a full connection neural network and training the full connection neural network, so as to obtain neural network model parameters; 2. The method of claim 1, wherein, The flight state sample space is obtained by cross combination of flight height H i , Mach number M j , and attack angle α k sample points, wherein the flight height interval is not more than 2 km, the Mach number interval is not more than 2, and the attack angle interval is not more than 2°, the flight height H i , Mach number M j , and attack angle α k are all positive integers.
3. The method of claim 2, wherein, constructing a neural network proxy model according to the neural network model parameters. The method comprises the following steps:
4. The method of claim 3, wherein the method further comprises: determining nose radius, leading edge radius, leading edge sweep angle and centerline cone angle geometric parameters according to the aircraft shape to be calculated, taking the geometric parameters as shape input of aerodynamic thermal engineering calculation, and performing aerodynamic thermal engineering calculation on all state points in the flight state sample space, so as to obtain thermal flow and transition intermittency factor data of all state points of the selected typical part of the aircraft. The method comprises the following steps:
5. The method of claim 4, wherein, the flight state sample point with the transition intermittency factor of 0 is a laminar flow state sample point, the flight state sample point with the transition intermittency factor of 1 is a turbulent flow state sample point, and the flight state sample point with the transition intermittency factor between 0 and 1 is a transition state sample point. The method comprises the following steps:
6. The method of claim 5, wherein the method further comprises: the data of the laminar flow state sample points are combined into a laminar flow state training data set, the data of the turbulent flow state sample points are combined into a turbulent flow state training data set, and the data of the transition state sample points are taken as a transition state training data set. The method comprises the following steps:
7. The method of claim 6, wherein the method further comprises: after the transition state sample space is encrypted, the flight height interval of the data in the transition state sample space is not more than 0.2 km, the Mach number interval is not more than 0.5, and the attack angle interval is not more than 0.5 degrees; the transition state sample space state points are calculated again by aerodynamic thermal engineering, so as to obtain the transition intermittency factor data of all state points in the encrypted transition state sample space. A four-layer fully connected neural network is constructed, wherein the input layer includes three variables: height, Mach number, and attack angle, the output layer is heat flux, and the two hidden layers each include 10 hidden variables; the flight state [H i , M j , α k ] in the laminar state sample space is input, and the heat flux is output; the fully connected neural network is trained by using a back propagation algorithm for the laminar state training data set, and the laminar state heat flux neural network model parameters are obtained.
8. The method of claim 7, wherein the method further comprises: The method comprises the following steps: A four-layer fully connected neural network is constructed, wherein the input layer includes three variables: height, Mach number, and attack angle, the output layer is heat flux, and each of the two hidden layers includes 10 hidden variables; flight states [H i , M j , α k ] in the sample space of the turbulent state are taken as inputs, and heat flux is taken as an output; the fully connected neural network is trained by using a back propagation algorithm for a training data set of the turbulent state, so as to obtain a neural network model parameter of the heat flux of the turbulent state.
9. The method of claim 8, wherein the method further comprises: The method comprises the following steps: A 4-layer fully connected neural network is constructed, wherein the input layer includes 3 variables: height, Mach number, and attack angle, the output layer is the intermittency factor, and each of the two hidden layers includes 10 hidden variables; flight states [H i , M j , α k ] in the transition state sample space are taken as inputs, and the intermittency factor is taken as an output; the neural network is trained by using a back propagation algorithm for the training data set of the transition intermittency factor, and the intermittency factor neural network model parameters are obtained.
10. The method of claim 8, wherein the method further comprises: The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: For any given new state [altitude H new , Mach number M new , angle of attack α new ] in the flight parameter range, the laminar flow heat flux Q L is calculated by using the laminar state heat flux neural network with the new state [H new , M new , α new ] as the input, the turbulent flow heat flux Q T is calculated by using the turbulent state heat flux neural network, and the intermittency factor γ is calculated by using the intermittency factor neural network, so as to take γ·Q L +(1-γ)·Q T as the neural network agent model heat flux output of the aerodynamic heat environment.