Airfoil profile optimization algorithm based on data driving
Through the data-driven airfoil optimization algorithm, deep neural network and sampling gradient algorithm are used to solve the problem of relying on experience and computational volume in the existing technology, and efficient airfoil optimization is achieved, which is suitable for all gradient algorithms and Avoid abnormal shapes.
Patent Information
- Application Number
- CN202411072172.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-06
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-06
AI Technical Summary
The existing airfoil optimization algorithm relies on engineer experience, has a long iteration cycle, is large in computing, is not versatile, and has low design efficiency.
Using data-driven airfoil optimization algorithm, the airfoil is used to sample the geometric feature space, and the airfoil is marked airfoil database is established, the agent model is trained using deep neural network, and combined with sampling gradient algorithms for optimization, which has low calculation cost and high efficiency. .
It greatly improves computing efficiency without sacrificing accuracy, has strong scalability, avoids shape abnormalities, and is suitable for all gradient algorithms, and has significantly improved optimization efficiency.
Smart Images

Figure CN119004971A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of CFD and artificial intelligence, in particular to an airfoil optimization algorithm based on data drive. Background Art
[0002] Airfoil optimization is a classic problem in aerodynamic design. Its goal is to obtain good aerodynamic performance and improve cruise efficiency by optimizing the airfoil shape.
[0003] The design algorithms disclosed in the prior art are usually obtained by gradually fine-tuning based on the design experience of engineers or by using the adjoint method. The manual fine-tuning method usually relies too much on the experience and level of engineers, and has a long iteration cycle and low design efficiency. The adjoint method obtains the relevant gradient by solving the adjoint equation of the corresponding PDE (partial differential equation), and then iterates using gradient-based optimization, but this method has a large amount of calculation and is not universal. Summary of the invention
[0004] In order to overcome the above-mentioned defects in the prior art, the present invention provides an airfoil optimization algorithm based on data-driven, which has low calculation cost and higher optimization efficiency.
[0005] To achieve the above object, the present invention adopts the following technical solutions, including:
[0006] A data-driven airfoil optimization algorithm includes the following steps:
[0007] S1, modeling the airfoil and sampling it in the geometric feature space to obtain the airfoil database to be labeled;
[0008] S2, using CAE software, solve the aerodynamic characteristics of the airfoil data to be annotated at different angles of attack, obtain the lift coefficient and the drag coefficient, and thus obtain an annotated airfoil database;
[0009] S3, using a deep neural network to train on a labeled airfoil database, with the airfoil geometry and angle of attack as input, and the lift coefficient and drag coefficient as output, to finally obtain a proxy model;
[0010] S4, sampling gradient algorithm, calculates the corresponding aerodynamic indicators and gradients for the optimized airfoil through the proxy model, and finally obtains the optimized airfoil.
[0011] Preferably, in step S1, the airfoil is modeled in the NACA-4 SERIES-MODIFIED format, and geometric features of the airfoil in five dimensions, namely, maximum radian, maximum radian position, maximum thickness, maximum thickness position and leading edge radius index, are extracted.
[0012] Preferably, the specific method of step S1 is:
[0013] S11, through the maximum thickness t max , maximum thickness position u max and leading edge radius index I to calculate the thickness curve:
[0014] First, solve two linear equations:
[0015]
[0016]
[0017] Where k(·) is a B-spline function obtained by interpolating the data {(0.20, 1.00), (0.30, 1.17), (0.40, 1.575), (0.500, 2.325), (0.6, 3.500)};
[0018] is an intermediate variable;
[0019]
[0020]
[0021] Among them, r LE is the radius of curvature of the leading edge; a0 is the coefficient of the first half of the thickness curve;
[0022] The calculation formula of the thickness curve t(u) is as follows:
[0023]
[0024] Where, u represents the horizontal coordinate value; t(u) represents the thickness at the current position;
[0025] S12, calculate the radian curve c(u) by the maximum radian M and the maximum radian position P. The calculation formula of the radian curve c(u) is as follows:
[0026]
[0027] Among them, c(u) represents the middle position of the current position;
[0028] S13, obtaining the airfoil profile by calculating the thickness curve and the camber curve, and retaining the airfoil parameters and airfoil profile data;
[0029] The airfoil profile scatter data is obtained by the following method:
[0030]
[0031] Among them, y upper (u) and ylower (u) represent the upper and lower contours of the airfoil, respectively.
[0032] Preferably, in step S2, the CAE software framework used is XFoil, which calculates the corresponding lift coefficient C of the airfoil at different angles of attack. l and the drag coefficient C d .
[0033] Preferably, the Mach number is 0.785 and the Reynolds number is 5×10 6 , the angle of attack ranges from -5° to 10°.
[0034] Preferably, in step S3, the network framework used is a fully connected neural network, and the activation function is ReLU.
[0035] Preferably, in step S4, the lift coefficient constraint is given The optimization goal is:
[0036]
[0037] Preferably, in step S4, the gradient algorithm used is the PSQP algorithm.
[0038] An electronic device comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned data-driven airfoil optimization algorithm when executing the computer program.
[0039] A computer program product includes a computer program / instruction, which implements the above-mentioned data-driven airfoil optimization algorithm when executed by a processor.
[0040] The advantages of the present invention are:
[0041] (1) The present invention provides an airfoil optimization algorithm based on data drive.
[0042] (2) The proxy model of the present invention greatly improves computational efficiency without sacrificing too much accuracy.
[0043] (3) The optimization algorithm of the present invention has strong scalability and can be applied to all gradient algorithms.
[0044] (4) Since the optimization is performed at the geometric feature space level, shape anomalies can be avoided. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 The present invention is a data-driven airfoil optimization algorithm flow chart.
[0046] Figure 2Schematic diagram of geometric parameters in this embodiment.
[0047] Figure 3 FIG. 4 is a schematic diagram of the network structure in this embodiment.
[0048] Figure 4 It is a schematic diagram of the optimization comparison results in this embodiment. DETAILED DESCRIPTION
[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0050] Example 1
[0051] Depend on Figure 1 As shown, a data-driven airfoil optimization algorithm specifically includes the following steps:
[0052] S1, model the airfoil in the NACA-4SERIES-MODIFIED format and perform sampling in the geometric feature space to obtain the airfoil database to be annotated.
[0053] S2, using CAE software, solve the aerodynamic characteristics of the airfoil data to be annotated at different angles of attack, obtain the lift coefficient and the drag coefficient, and thus obtain a labeled airfoil database.
[0054] S3, uses a deep neural network to train on a labeled airfoil database. The input is the geometric features of the five dimensions of the airfoil data and the angle of attack. The output is the lift coefficient and the drag coefficient. Finally, the proxy model is obtained.
[0055] S4, sampling gradient algorithm, calculates the corresponding aerodynamic indicators and gradients for the optimized airfoil through the proxy model, and finally obtains the optimized airfoil.
[0056] In step S1, the airfoil is modeled in the NACA-4SERIES-MODIFIED format. The geometric features of the airfoil include the maximum radian M, the maximum radian position P, the maximum thickness t max , maximum thickness position u max and leading edge radius index r LE The geometric parameters of the airfoil are the geometric characteristics of the five dimensions. Figure 2 As shown, the middle blue line represents the radian curve c(u).
[0057] S11, through the maximum thickness t max, maximum thickness position u max The thickness curve is calculated by solving two linear equations:
[0058]
[0059]
[0060] Where k(·) is a B-spline function obtained by interpolating the data {(0.20, 1.00), (0.30, 1.17), (0.40, 1.575), (0.500, 2.325), (0.6, 3.500)};
[0061] is an intermediate variable;
[0062]
[0063]
[0064] Among them, r LE is the curvature radius of the leading edge; a0 is a known coefficient of the first half of the thickness curve, which can be converted into t by the leading edge radius index I and the maximum thickness max It is also used to assist in calculating other unknown coefficient values a1, a2, and a3 in the first half of the thickness curve.
[0065] The calculation formula of the thickness curve t(u) is as follows:
[0066]
[0067] Wherein, u represents the abscissa value, and t(u) represents the thickness at the current position;
[0068] S12, calculating the radian curve c(u) by the maximum radian M and the maximum radian position P. The calculation formula of the radian curve c(u) is as follows:
[0069]
[0070] Where c(u) represents the middle position of the current position, see Figure 2 The middle blue curve of
[0071] S13, obtain the contour of the airfoil curve by calculating the thickness curve and the camber curve, and retain the airfoil parameters and airfoil contour data. The contour scatter data of the airfoil is obtained by the following method:
[0072]
[0073] Among them, y upper(u) and y lower (u) represent the upper and lower contours of the airfoil, respectively.
[0074] The airfoil data to be annotated are recorded as the geometric feature data of the above five dimensions, and the airfoil profile can be calculated based on the geometric features of the five dimensions.
[0075] In step S2, the CAE software framework used is XFoil, which is an open source software for calculating airfoils. The method is based on the panel method to calculate the corresponding lift coefficient C of the airfoil at different angles of attack. l and the drag coefficient C d , where the Mach number is 0.785 and the Reynolds number is 5×10 6 , the angle of attack ranges from -5° to 10°. CAE software solves the aerodynamic characteristics of the airfoil at different angles of attack based on the airfoil profile data to obtain the lift coefficient and drag coefficient.
[0076] Depend on Figure 3 As shown, in step S3, the network framework used is a fully connected neural network, the activation function is ReLU, the input is the geometric features of the five dimensions of the airfoil data and the angle of attack, the output is the lift coefficient and the drag coefficient, and finally the proxy model is obtained. The network structure is shown in Table 1 below:
[0077] Table 1 Structure of deep neural network
[0078] Linear(6,128) ReLU Linear(128,128) ReLU Linear(128,128) ReLU Linear(128,128) ReLU Linear(128,2)
[0079] In step S4, the gradient algorithm used is the PSQP algorithm (sequential quadratic programming algorithm), and the specific method is as follows:
[0080] S41, given lift coefficient constraint Solve the following optimization problem:
[0081]
[0082] S42, for the airfoil to be optimized, the corresponding aerodynamic index, namely the lift coefficient C, is calculated through the proxy model. l and the drag coefficient C d ;
[0083] S43, the gradient calculation required by the optimization algorithm is obtained by automatically differentiating the proxy model. The gradient is used to indicate the optimization direction; the optimization algorithm needs to calculate the optimization direction based on the gradient and iteratively update according to the optimization direction; Pytorch's network model has an automatic differentiation mechanism.
[0084] The optimized airfoil is as follows Figure 4 As shown, Figure 4The black curve in the middle is the original airfoil, and the purple curve is the airfoil optimized according to the target parameters (angle of attack, Mach number, etc.).
[0085] Example 2
[0086] An electronic device comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the processor implements a data-driven airfoil optimization algorithm of the above-mentioned embodiment 1 when executing the computer program.
[0087] The electronic device of the embodiment of the present application can be the mobile device itself, or a stand-alone device independent of the mobile device, which can communicate with the mobile device to receive the collected input signals from them and send the selected target decision behavior to them.
[0088] The electronic device includes one or more processors and memory. The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may run the program instructions to implement the decision behavior decision method and / or other desired functions of the various embodiments of the present application described above.
[0089] Electronic devices may also include input devices and output devices.
[0090] Example 3
[0091] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the decision-making behavior decision method according to various embodiments of the present application described in the above-mentioned embodiment 1 of this specification.
[0092] The computer program product may be written in any combination of one or more programming languages to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.
[0093] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A data-driven airfoil optimization algorithm, characterized in that: The following steps are involved: S1, modeling the airfoil and sampling it in the geometric feature space to obtain the airfoil database to be labeled; S2, using CAE software, solve the aerodynamic characteristics of the airfoil data to be annotated at different angles of attack, obtain the lift coefficient and the drag coefficient, and thus obtain an annotated airfoil database; S3, using a deep neural network to train on a labeled airfoil database, with the airfoil geometry and angle of attack as input, and the lift coefficient and drag coefficient as output, to finally obtain a proxy model; S4, sampling gradient algorithm, calculates the corresponding aerodynamic indicators and gradients for the optimized airfoil through the proxy model, and finally obtains the optimized airfoil.
2. The data-driven airfoil optimization algorithm according to claim 1, characterized in that: In step S1, the airfoil is modeled in the NACA-4 SERIES-MODIFIED format, and geometric features of the airfoil in five dimensions, namely, maximum camber, maximum camber position, maximum thickness, maximum thickness position, and leading edge radius index, are extracted.
3. The data-driven airfoil optimization algorithm according to claim 2, characterized in that: The specific method of step S1 is: S11, through the maximum thickness t max , maximum thickness position u max and leading edge radius index I to calculate the thickness curve: First, solve two linear equations: Where k(·) is a B-spline function obtained by interpolating the data {(0.20, 1.00), (0.30, 1.17), (0.40, 1.575), (0.500, 2.325), (0.6, 3.500)}; is an intermediate variable; Among them, r LE is the radius of curvature of the leading edge; a0 is the coefficient of the first half of the thickness curve; The calculation formula of the thickness curve t(u) is as follows: Where, u represents the horizontal coordinate value; t(u) represents the thickness at the current position; S12, calculate the radian curve c(u) by the maximum radian M and the maximum radian position P. The calculation formula of the radian curve c(u) is as follows: Among them, c(u) represents the middle position of the current position; S13, obtaining the airfoil profile by calculating the thickness curve and the camber curve, and retaining the airfoil parameters and airfoil profile data; The airfoil profile scatter data is obtained by the following method: Among them, y upper (u) and y lower (u) represent the upper and lower contours of the airfoil, respectively.
4. The data-driven airfoil optimization algorithm according to claim 1, characterized in that: In step S2, the CAE software framework used is XFoil, which calculates the corresponding lift coefficient C of the airfoil at different angles of attack. l and the drag coefficient C d .
5. The data-driven airfoil optimization algorithm according to claim 4, characterized in that: The Mach number is 0.785 and the Reynolds number is 5×10 6 , the angle of attack ranges from -5° to 10°.
6. The data-driven airfoil optimization algorithm according to claim 1, characterized in that: In step S3, the network framework used is a fully connected neural network, and the activation function is ReLU.
7. The data-driven airfoil optimization algorithm according to claim 1, characterized in that: In step S4, the lift coefficient constraint is given The optimization goal is:
8. The data-driven airfoil optimization algorithm according to claim 1, characterized in that: In step S4, the gradient algorithm used is the PSQP algorithm.
9. An electronic device, characterized in that: It comprises a processor, a memory and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, an airfoil optimization algorithm based on data-driven according to any one of claims 1 to 8 is implemented.
10. A computer program product, characterized in that It includes a computer program / instruction, which, when executed by a processor, implements a data-driven airfoil optimization algorithm as described in any one of claims 1 to 8.
Citation Information
Patent Citations
Airfoil steady aerodynamic optimization design method considering geometric uncertainty factors
CN111737928A
Variant aircraft aerodynamic optimization method based on improved position vector expectation improvement degree
CN112329140A
Deep neural network modeling method for large-difference aerodynamic data
CN113836634A
Miniature solar unmanned aerial vehicle layered pneumatic optimization method
CN116644516A
Aircraft robust optimization design method based on gradient enhanced random Co-Kriging model
CN117077298A
Cited By
Pneumatic correction method and device for wind turbine airfoil and medium
CN120046493A
