CN108733914A - Transonic airfoil Natural Laminar Flow delay based on artificial neural network turns to twist design method - Google Patents

Abstract

The invention belongs to the technical field of aircraft design, and specifically relates to a method for designing a natural laminar flow delay transition of a transonic airfoil based on an artificial neural network. The specific steps of the present invention are as follows: (1) use the airfoil parameterization method to analyze the shape of the airfoil, and establish the airfoil parameterization expression; (2) according to the aerodynamic characteristics of the airfoil in the transonic cruise flow field, extract and transonic Aerodynamic parameters related to natural laminar flow at the speed of sound; (3) Using artificial neural network technology to realize intelligent system optimization; finally output corresponding new airfoil data. The method of the invention can realize the expansion of the laminar flow area on the surface of the airfoil in the complex flow environment of the (weak) shock wave-boundary layer under the transonic cruise flight of the aircraft, thereby delaying the occurrence of the laminar flow transition and realizing the reduction of frictional resistance small goals. The new airfoil with excellent transonic natural laminar flow performance obtained by the invention can be applied to the design optimization of aircraft wings and engine nacelle sections.

Description

Delayed transition design method for natural laminar flow of transonic airfoil based on artificial neural network

technical field

The invention belongs to the technical field of aircraft design, and in particular relates to a design method for a natural laminar flow delay transition of an aircraft transonic airfoil.

Background technique

The design of modern civil aircraft needs to meet the "four characteristics", namely safety, economy, comfort and environmental protection. Under the situation that European and American countries already have absolute advantages in this field, the international competition is very fierce, and the market entry threshold is very high, we need to pay attention to basic research, establish strong scientific research strength and long-term technical reserves, and form independent innovation and Sustainability.

Aircraft drag directly determines aircraft economy and emissions. Drag reduction is closely related to economy and environmental protection, among which economy is the main factor that determines the airline's aircraft purchase, and environmental protection may become the access condition of the future aviation market. The flow field of a large passenger aircraft is an important factor determining the aerodynamic performance of the aircraft. The transition from laminar flow to turbulent flow is due to boundary layer instability and further growth of disturbances, while boundary layer instability originates from small velocity disturbances that grow or decay, and the growth of small velocity disturbances leads to large amplitude, non-linear properties, and eventually the flow transitions from laminar to turbulent. In the case of linear non-separation, the energy of the fluid is consumed in the shearing motion in the critical region, and in the case of separation, a larger part of the energy is lost in the vortex motion after the edge separation.

The distribution of laminar flow and turbulent flow in the flow field on the surface of the aircraft plays a more important role. For example, the flow field resistance of a large airliner flight includes frictional resistance, induced resistance, and shock wave resistance, etc., delaying the occurrence of transition as much as possible, that is, expanding the area of laminar flow on the surface of the model, reducing the area of turbulent flow infiltration, and reducing the flow control. Effective means of frictional resistance. As another example, the reduction of the turbulent flow area will reduce the turbulent flow excitation noise. These examples all illustrate the key role of the distribution of laminar/turbulent flow on the aerodynamic performance of the aircraft, and also illustrate the urgent need for the delayed transition technology of the aircraft wing.

At present, the design of conventional aircraft is already at the bottleneck stage, and advanced theories and methods are bringing profound changes and progress to the development of large passenger aircraft. In order to improve the "four characteristics" of civil aircraft, Europe, the United States and other countries have launched a series of prospective study. Europe has put forward the so-called "green aircraft challenge" in its civil aviation 2020 long-term plan. The main goals include: reducing NOX emissions by 80%, CO2 emissions by 50% (fuel consumption is halved), and accident probability by 2020. Aircraft can feel that the noise is halved (the noise at the airport boundary is reduced to 65dB), and the turnover efficiency of aviation operations is greatly improved. After NASA implemented the Advanced Subsonic Aircraft Technology (AST) research program from 1994 to 2001, it launched the Quiet Aircraft Technology (QAT) research program in 2001. It plans to further reduce aircraft noise by 5dB compared with the level in 1997. The future goal is 20dB reduction in 25 years.

In order to enable my country to form a large passenger aircraft industry with international competitiveness as soon as possible, it is very necessary to strengthen research on relevant basic scientific issues, consolidate the foundation for the development of large passenger aircraft, develop new research methods, build independent innovation capabilities, and achieve sustainable development in core technologies. Development, serving the needs of reality and future aviation.

Contents of the invention

The object of the present invention is to provide a kind of aircraft transonic airfoil natural laminar delay transition design method, make under the aircraft transonic cruise flight, in the complex flow environment of (weak) shock wave-boundary layer, realize airfoil surface The expansion of the laminar flow area delays the occurrence of laminar flow transition and achieves the goal of reducing frictional resistance.

The aircraft transonic airfoil natural laminar delay transition design method proposed by the present invention is based on artificial neural network technology, and the specific steps are as follows:

(1) Analyze the shape of the airfoil by using the airfoil parameterization method, and establish the parametric expression method of the airfoil;

Using the PARSEC parameterization method (Sobieczky H.Parametric airfoils and wings[M]//Recent Development of Aerodynamic Design Methodologies. Vieweg+Teubner Verlag, 1999:71-87.) to describe the aircraft airfoil and establish the expression of the aircraft airfoil, namely Take the following 11 PARSEC parameters: leading edge radius r _le , maximum thickness X _up and X _lo of the upper/lower airfoil, corresponding positions Z _up and Z _lo of the maximum thickness of the upper/lower airfoil, apex curvature Z _xxup of the upper/lower airfoil and Z _xxlo , trailing edge width △ Z _TE , trailing edge vertical height Z _TE , trailing edge wedge angle β _TE , trailing edge direction angle α _TE , to simulate airfoil geometry. The PARSEC parameterization method succinctly associates the flow characteristics with the airfoil geometric characteristics, which is of great benefit to both design and research. Therefore, the parametric expressions of the upper and lower airfoil curves are shown in formula (1):

a _n (n=1,2,...,6) are polynomial coefficients. For the upper airfoil, the coefficient a _n is given by the matrix equation (2):

For the lower airfoil, the coefficient a _n is given by the matrix equation (3), which is similar to the upper airfoil, that is, the parameters representing the configuration characteristics of the upper airfoil are replaced by the corresponding parameters of the lower airfoil:

By obtaining the fitting coefficient, the relationship between the geometric parameters and the actual airfoil shape can be established. Where x _up is the abscissa of each point on the upper airfoil, and x _te is the abscissa of the trailing edge of the airfoil.

(2) According to the aerodynamic characteristics of the airfoil in the flow field of transonic cruise, the aerodynamic parameters related to the transonic natural laminar flow are extracted; the flow field calculation equation is used to calculate the flow field and obtain the result of the length of the laminar flow region. The NS equation (Currie, IG (1974), Fundamental Mechanics of Fluids, McGraw-Hill, ISBN 0-07-015000-1) was used to solve the flow field, and the SST turbulence model (Menter, FR (August 1994), "Two-EquationEddy -Viscosity Turbulence Models for Engineering Applications", AIAA Journal, 32(8):1598–1605, Bibcode:1994AIAAJ..32.1598M, doi:10.2514/3.12149). When solving the laminar-turbulent flow distribution, the γ-Re transition model based on the intermittent factor of the SST turbulence model (Langtry R, MenterF. Transition Modeling for General CFD Applications in Aeronautics [M]. 2015.), as shown in formula (4 )(5), where ρ is the density, U is the velocity, p is the pressure, γ is the turbulent intermittent factor, Re _θt is the local momentum Reynolds number, μ is the molecular viscosity, μ _t is the turbulent viscosity assumed by Boussinesq, P _γ is the generation item of the turbulent intermittent factor, E _γ is the dissipation item of the turbulent intermittent factor, U _j is the velocity in the j direction (j=1, 2, 3), σ _f is the coefficient calibrated by the experiment, P _θt is the momentum Reynolds number Produced items:

The flow field solution is obtained by solving the above equations, and the corresponding aerodynamic performance parameters of the length of the laminar flow region are obtained.

(3) Use artificial neural network technology to realize intelligent system optimization; finally output corresponding new airfoil data, and the new airfoil has improved transonic natural laminar flow characteristics.

The optimal design is realized through the artificial neural network (ANN) algorithm (Zeidenberg M. Neural networks in artificial intelligence [M]. Ellis Horwood, 1990.). Regarding the training method of the neural network, the GRNN algorithm (Specht D F.A general regression neural network. Neural Networks[J], IEEE Transactions on, 1991.2(6): p.568-576.) was used (after comparison, the level of the target correlation coefficient and network generalization ability are better than BP algorithm and RBF algorithm). Generalized regression neural network (GRNN) was proposed by American scholar Donald F. Specht in 1991. It is a variation of radial basis neural network. GRNN is a feed-forward neural network based on nonlinear regression analysis. It has strong nonlinear mapping ability, flexible network structure, and high fault tolerance and robustness. It is suitable for solving nonlinear problems. GRNN consists of four layers. , are the input layer, model layer, summation layer and output layer, respectively.

The optimization process here is an iterative process. In one optimization step, firstly, according to the initial airfoil to be optimized, the corresponding 11 PARSEC geometric parameters are analyzed; through the Latin hypercube sampling technique (Owen, A.B. (1992). "Orthogonal arrays for computer experiments, integration and visualization" .Statistica Sinica.2:439–452.) Perform parameter perturbation to obtain a series of new airfoil corresponding geometric parameters. The perturbation range is gradually reduced in each optimization step in order to eventually converge to an optimized transonic laminar airfoil. For a series of airfoils obtained by parameter perturbation, the flow field calculation equation mentioned above is used to calculate the flow field and obtain the result of the length of the laminar flow region.

At each optimization step, each new airfoil has corresponding geometric parameters and aerodynamic performance parameters for the length of the laminar flow region. This method uses an agent model based on artificial neural network to establish the connection between the two. When training the artificial neural network, the input end is the aerodynamic performance (laminar flow length) of each airfoil in this optimization step, and the output end is the geometric parameters of each airfoil. Then train to obtain a proxy model.

Next, the optimization of this optimization step is carried out, and an appropriate target laminar region length parameter is input into the trained artificial neural network, in order to obtain the geometric parameters corresponding to the new airfoil. "Appropriate" here mainly means: 1. The length of the input target laminar flow area should not be too ideal, because it does not conform to our concept of step-by-step optimization; 2. Too high input will cause the shape of the new generated airfoil to be distorted .

A suitable input generally refers to the best aerodynamic performance among a series of airfoils generated by parameter perturbation in this optimization step. As the iteration proceeds, the input parameters of the appropriate target laminar flow region length are gradually optimized, and finally converge to a parameter corresponding to the optimal transonic natural laminar flow airfoil.

Beneficial effects created by the present invention:

The method of the invention can realize the enlargement of the laminar flow area on the surface of the airfoil in the complex flow environment of (weak) shock wave-boundary layer under the transonic cruising flight of the aircraft, thereby delaying the occurrence of laminar flow transition and realizing the reduction of frictional resistance small goals. The obtained new airfoil with excellent transonic natural laminar flow performance can be applied to the design optimization of aircraft wing and engine nacelle section. Figure 5 and Figure 6 respectively show the fact that the length of the laminar region on the surface of the transonic natural laminar flow airfoil before and after the optimization obtained by the experiment increases (the highlighted part is the laminar flow region, and the boundary between the bright part and the relatively dark part is the laminar transition zone). When the angle of attack is 1°, the length of the laminar flow region of the initial airfoil before optimization is 69.2% of the unit chord length, and the length of the transonic natural laminar flow region after optimization is 76.8% of the unit chord length; when the angle of attack is 4°, the layer of the initial airfoil before optimization The flow region length is 60.1% of the unit chord length, and the optimized transonic natural laminar flow region length is 63.5% of the unit chord length.

Description of drawings

Figure 1 is a comparison of the laminar-turbulent flow distribution in the airfoil surface flow before and after optimization, illustrating the design intent of the transonic laminar flow airfoil.

Figure 2 is a schematic diagram of the flow field mesh divided during the calculation of the airfoil flow field.

Fig. 3 is a schematic diagram of the design optimization method of transonic natural laminar flow airfoil based on artificial neural network.

Figure 4 is a graph comparing the accuracy of the transition model used with the wind tunnel test results, illustrating the reliability of the flow field simulation solver used in determining the length of the laminar flow region.

Figure 5 is a comparison of the non-transonic natural laminar flow airfoil (upper) and the optimized transonic natural laminar flow airfoil (lower) at Mach number 0.785 in the laminar flow region with an angle of attack of 1°.

Figure 6 is a comparison of the non-transonic natural laminar flow airfoil (upper) and the optimized transonic natural laminar flow airfoil (lower) at a Mach number of 0.785 and a laminar flow region with an angle of attack of 4°.

Fig. 7 is a flow chart diagram of the optimization design of the present invention.

Detailed ways

Step (1): Establish an airfoil parameterization method for optimal design of transonic laminar flow.

The PARSEC parameterization method is used to express the shape of the airfoil, and the expression method of the aircraft airfoil is established, see Table 1.

Table 1 Geometric parameters of an initial airfoil before optimization

Step (2) parameter perturbation to obtain a series of airfoils

The optimization process here is an iterative process. In one optimization step, firstly, according to the initial airfoil to be optimized, the corresponding 11 PARSEC geometric parameters are analyzed; through the Latin hypercube sampling technique (Owen, A.B. (1992). "Orthogonal arrays for computer experiments, integration and visualization" .Statistica Sinica.2:439–452.) Perform parameter perturbation to obtain a series of new airfoil corresponding geometric parameters. The perturbation range is gradually reduced in each optimization step in order to eventually converge to an optimized transonic laminar airfoil. For a series of airfoils obtained by parameter perturbation, the flow field calculation equation mentioned above is used to calculate the flow field and obtain the result of the length of the laminar flow region.

At each optimization step, each new airfoil has corresponding geometric parameters and aerodynamic performance parameters for the length of the laminar flow region. This method uses an agent model based on artificial neural network to establish the connection between the two. When training the artificial neural network, the input end is the aerodynamic performance (laminar flow length) of each airfoil in this optimization step, and the output end is the geometric parameters of each airfoil. Then perform training to obtain a proxy model. Step (3): Establish a training-related artificial neural network.

The new airfoils have corresponding geometric parameters and aerodynamic performance parameters for the length of the laminar flow region. The agency model is used to establish the connection between the two. Similar to the traditional inverse design, a suitable target laminar region length parameter is input into the trained artificial neural network in order to obtain the geometric parameters corresponding to the new airfoil. Refer to Figure 3 for the design optimization idea of this part.

With the iteration of the optimization step, the input parameters of the appropriate target laminar flow region length are gradually optimized, and finally converge to a parameter corresponding to the optimal transonic natural laminar flow airfoil. The entire optimization process can be seen in Figure 7. Figure 5 and Figure 6 respectively show the fact that the length of the laminar region on the surface of the transonic natural laminar flow airfoil before and after the optimization obtained by the experiment increases (the highlighted part is the laminar flow region, and the boundary between the bright part and the relatively dark part is the laminar transition zone). When the angle of attack is 1°, the length of the laminar flow region of the initial airfoil before optimization is 69.2% of the unit chord length, and the length of the transonic natural laminar flow region after optimization is 76.8% of the unit chord length; when the angle of attack is 4°, the layer of the initial airfoil before optimization The flow region length is 60.1% of the unit chord length, and the optimized transonic natural laminar flow region length is 63.5% of the unit chord length.

Claims (2)

1. a kind of aircraft transonic airfoil natural laminar delay transition design method, is based on artificial neural network technology, is characterized in that, concrete steps are as follows: (1) Use the airfoil parameterization method to analyze the shape of the airfoil, and establish the parametric expression method of the airfoil: Using the PARSEC parameterization method to describe the aircraft airfoil, establish the expression of the aircraft airfoil, that is, use the following 11 PARSEC parameters: leading edge radius r _le , maximum thickness X _up and X _lo of the upper/lower airfoil, maximum upper/lower airfoil Thickness corresponds to position Z _up and Z _lo , curvature of upper/lower airfoil apex Z _xxup and Z _xxlo , trailing edge width △ Z _TE , trailing edge vertical height Z _TE , trailing edge wedge angle β _TE , trailing edge direction angle α _TE , Simulate the geometry of the airfoil; thus, the parametric expressions of the upper and lower airfoil curves are shown in formula (1): (1) a _n is the polynomial coefficient, n =1,2,…,6; for the upper airfoil, the coefficient a _n is given by the matrix equation (2): (2) For the lower airfoil, the coefficient a _n is given by the matrix equation (3): (3) Obtain the fitting coefficient, that is, establish the connection between the geometric parameters and the actual airfoil shape; where x _up is the abscissa of each point on the upper airfoil, and x _te is the abscissa of the trailing edge of the airfoil; (2) According to the aerodynamic characteristics of the airfoil in the transonic cruise flow field, extract the aerodynamic parameters related to the transonic natural laminar flow: use the flow field calculation equation to calculate the flow field and obtain the result of the length of the laminar flow region; solve the flow field Select the NS equation and use the SST turbulence model; when solving the laminar-turbulent flow distribution, apply the γ-Re transition model based on the intermittent factor of the SST turbulence model, as shown in formula (4) (5), where ρ is the density, U is the velocity, p is the pressure, γ is the turbulent intermittent factor, Re _θt is the local momentum Reynolds number, μ is the molecular viscosity, μ _t is the turbulent viscosity according to Boussinesq assumption, P _γ is the turbulent intermittent factor generating term, E _γ is the turbulent flow Intermittent factor dissipation item, U _j is the velocity in the j direction, j=1, 2, 3, σ _f is the coefficient calibrated by the experiment, P _θt is the momentum Reynolds number generation item: (4) (5) The flow field solution is obtained by solving the above equations, and the corresponding aerodynamic performance parameters of the length of the laminar flow region are obtained; (3) Use artificial neural network technology to realize intelligent system optimization; finally output the corresponding new airfoil data. 2. The design method according to claim 1, characterized in that the optimization described in step (3) is an iterative process, and the specific process is: In each optimization step, firstly, according to the initial airfoil to be optimized, the corresponding 11 PARSEC geometric parameters are analyzed; the parameters are perturbed through the Latin hypercube sampling technique, and a series of new airfoil corresponding geometric parameters are obtained; For a series of airfoils obtained by parameter perturbation, the flow field calculation equation in step (2) is used to calculate the flow field and obtain the result of the length of the laminar flow region; In each optimization step, each new airfoil has corresponding geometric parameters and aerodynamic performance parameters of the length of the laminar flow region; a proxy model based on artificial neural network is used to establish the connection between the two; when training the artificial neural network , the input end is the aerodynamic performance of each airfoil in this optimization step-the laminar flow length, the output end is the geometric parameters of each airfoil, and then train to obtain the proxy model; Next, the optimization of this optimization step is carried out, and an appropriate target laminar flow region length parameter is input in the trained artificial neural network to obtain the geometric parameters corresponding to the new airfoil; the appropriate input refers to the parameter set in this optimization step The best aerodynamic performance in a series of airfoils generated by the disturbance; as the iteration proceeds, the input appropriate target laminar flow region length parameters are gradually optimized, and finally converge to a parameter corresponding to the optimal transonic natural laminar flow airfoil .