Transsonic shock wave buffeting suppression air blowing and suction flow channel design method based on deep learning

Abstract

The invention provides a transonic shock wave buffeting suppression blowing and sucking flow channel design method based on deep learning. A CST parameterization method, a TensorFlow-based multi-layer perceptron neural network model, a PSO optimization algorithm and the like are combined. In the CST parameterization process, the shape of the complex flow channel is accurately described with the minimum design variable, airfoil aerodynamic parameters corresponding to a large number of flow channel configurations generated randomly are predicted based on the built neural network model, and the prediction precision is high. And finally, constructing a target optimization function taking the maximum lift coefficient average value and the minimum lift coefficient fluctuation amplitude of the blowing and suction air flow channel as optimization targets, and performing global optimization in the constructed flow channel parameter database through a PSO algorithm to obtain optimal blowing and suction air flow channel structure parameters and the corresponding airfoil lift coefficient average value and amplitude. The optimized flow channel shows a good control effect on shock wave buffeting under the working conditions of different Mach numbers and attack angles, and the aerodynamic performance and the flow field characteristic of the blowing and sucking airfoil profile are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of turbulent boundary layer flow control, and particularly relates to a design method for a blowing and suction air duct for suppressing transonic shock buffet based on deep learning. Background Technique

[0002] Transonic shock buffet is a key aerodynamic problem frequently encountered by large transport aircraft and military aircraft in the transonic region. Under a specific combination of Mach number and angle of attack, the interaction between the shock wave and the boundary layer will trigger low-frequency and large-amplitude shock oscillations, thereby generating significant unsteady aerodynamic loads. Traditional blowing and suction control methods rely on empirical parameter adjustment and a large number of CFD simulation calculations, which have problems such as low optimization efficiency, high computational cost, and difficulty in achieving global optimality; moreover, existing blowing and suction air duct designs mostly use manual trial and error or simple parameterization methods, and are unable to efficiently handle complex geometric shapes and multi-objective optimization problems.

[0003] Therefore, there is an urgent need to design a new blowing and suction air duct for suppressing transonic shock buffet, which can suppress shock buffet in the transonic state, improve the aerodynamic performance of the airfoil, and significantly reduce the consumption of computing resources at the same time. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies of the cumbersome calculation process of the blowing and suction control method in the existing transonic shock buffet and the problem that the blowing and suction air duct structure design is not applicable to complex geometric shapes, and provides a design method for a blowing and suction air duct for suppressing transonic shock buffet based on deep learning.

[0005] To achieve the above purpose, the technical solution provided by the present invention is:

[0006] A design method for a blowing and suction air duct for suppressing transonic shock buffet based on deep learning, the blowing and suction air duct includes a suction port, a blowing port, and a duct provided between the suction port and the blowing port, the duct includes a left wall surface and a right wall surface, and the design method includes the following steps:

[0007] Step 1: Obtain the original coordinates of the duct wall surface, describe the geometric shape of the blowing and suction air duct using the CST parameterization method, construct the shape functions of the left wall surface and the right wall surface, determine the optimization variables of the blowing and suction air duct and perform variable parameter adjustment, and randomly generate multiple configurations of the blowing and suction air duct to obtain the structural parameters of the blowing and suction air duct;

[0008] Step 2: Mesh the internal flow field of some of the blowing and suction air ducts randomly generated in Step 1, and perform flow field calculations to obtain the aerodynamic parameters of the airfoil corresponding to different configurations of the blowing and suction air ducts, and establish a blowing and suction air duct database including the structural parameters of the blowing and suction air ducts and the corresponding aerodynamic parameters of the airfoil;

[0009] Step 3: Construct a multi-layer perceptron neural network model based on TensorFlow and perform training and prediction; the input of the multi-layer perceptron neural network model is the structural parameters of the blowing and suction air flow channel, and the output is the mean value and amplitude of the airfoil lift coefficient;

[0010] Step 4: Input the remaining structural parameters of the blowing and suction air flow channel in Step 1 into the multi-layer perceptron neural network model trained and predicted in Step 3, and output the corresponding mean value and amplitude of the airfoil lift coefficient;

[0011] Step 5: Based on the blowing and suction air flow channel database established in Step 2, as well as the structural parameters of the blowing and suction air flow channel and the corresponding mean value and amplitude of the airfoil lift coefficient obtained in Step 4 based on the multi-layer perceptron neural network model, construct an optimized database of the structural parameters of the blowing and suction air flow channel;

[0012] Step 6: Establish an objective optimization function with the average value of the maximum lift coefficient and the fluctuation amplitude of the minimum lift coefficient of the blowing and suction air flow channel as the optimization objectives, and use the PSO optimization algorithm to perform global optimization in the optimized database of the structural parameters of the blowing and suction air flow channel constructed in Step 5, and solve to obtain the optimal structural parameters of the blowing and suction air flow channel and the corresponding mean value and amplitude of the airfoil lift coefficient;

[0013] The objective optimization function is: minf val =-C L +ΔC L

[0014] In the formula: f val is the objective function value, minf val represents the comprehensive performance index to be minimized, C L is the normalized average lift coefficient, and ΔC L is the normalized lift coefficient fluctuation amplitude.

[0015] Furthermore, Step 1 includes the following sub-steps:

[0016] Step 1.1: Obtain the original coordinates (x, y) of the left and right walls of the blowing and suction air flow channel, and perform normalization processing to map the original wall coordinates of the blowing and suction air flow channel to the interval [0, 1];

[0017] The normalization processing formula is:

[0018]

[0019] where, x i is the dimensionless representation of the wall coordinate of the flow channel, x is the original wall coordinate of the flow channel, min(x) is the minimum value of the original wall coordinate of the flow channel, and max(x) is the maximum value of the original wall coordinate of the flow channel;

[0020] Step 1.2: Establish the parametric description equations of the left wall surface and the right wall surface using the CST parametric method.

[0021] The parametric description equation of the left wall surface is:

[0022] Y left (x i ) = C(x i )S left (x i ) + x i Y LE

[0023] The parametric description equation of the right wall surface is:

[0024] Y right (x i ) = C(x i )S right (x i ) + x i Y RE

[0025] In the formula, C(x i ) is a category function, and S left (x i ) and S right (x i ) are the left wall surface shape function and the right wall surface shape function respectively, which are constructed by Bernstein polynomials; Y LE and Y RE are the x - coordinates of the left wall surface and the right wall surface after normalization respectively, where the coordinate origin is located at the leading edge of the airfoil, the x - axis is along the chord direction of the airfoil, and the y - axis is along the longitudinal direction of the airfoil.

[0026] Step 1.3: Construct the shape function, determine the flow channel optimization variables and the flow channel geometric constraint conditions.

[0027] Construct the shape function: Select the weight coefficients of the 6 - order sub - shape function as the flow channel design parameters, and use Bernstein polynomials to construct the left wall surface shape function and the right wall surface shape function; The expression of the Bernstein polynomial S i (x) is:

[0028]

[0029] In the formula, B k,6 (x i ) is the 6 - order Bernstein basis function, and w k is the weight coefficient.

[0030] Select the position coordinates of the blowing port and the suction port of the blowing - suction flow channel as the flow channel geometric constraint conditions; and select several component shape functions as the flow channel optimization variables.

[0031] Step 1.4: Generate a parametric program for the runner design, adjust the variable parameters of the runner optimization variables, and randomly generate multiple blowing and suction runner configurations.

[0032] Further, in the step 1.3, five component shape functions are respectively selected at corresponding positions on the left wall surface and the right wall surface as the runner parameter optimization variables.

[0033] Further, in the step 1.4, generate a parametric program for the runner design based on MATLAB, adjust the variable parameters of the runner optimization variables; and randomly generate multiple blowing and suction runner configurations based on the LHS method.

[0034] Further, the step 2 includes the following sub-steps:

[0035] Step 2.1: Perform grid division on the internal flow field of some randomly generated blowing and suction runners in step 1 based on Pointwise software to generate a number of grids;

[0036] Step 2.2: Import the generated grids into the Fluent simulation software for flow field calculation to obtain the airfoil aerodynamic parameters corresponding to different configurations of the blowing and suction runners, where the airfoil aerodynamic parameters include the lift coefficient and the drag coefficient;

[0037] Step 2.3: Establish a blowing and suction runner database based on MATLAB, where the blowing and suction runner database includes the structural parameters of different configurations of the blowing and suction runners, as well as the corresponding airfoil lift coefficient and drag coefficient.

[0038] Further, the step 3 includes the following sub-steps:

[0039] Step 3.1: Construct a multi-layer perceptron neural network model based on TensorFlow, where the multi-layer perceptron neural network model includes an input layer, a fully connected layer, a batch normalization layer, an activation layer, a Dropout layer, and an output layer;

[0040] Set the hyperparameters of the multi-layer perceptron neural network model;

[0041] Step 3.2: Divide the dataset included in the blowing and suction runner database established in step 2 into a training set and a test set;

[0042] Step 3.3: Use the training set data to train the multi-layer perceptron neural network model, and perform real-time verification on the model during the training process, and dynamically adjust the hyperparameters of the multi-layer perceptron neural network model;

[0043] Step 3.4: Use the test set to predict the trained multi-layer perceptron neural network model.

[0044] Further, in step 3.1, the multi-layer perceptron neural network model adopts the Adam optimizer; the hyperparameters include the initial learning rate, batch size, and number of iterations, and the initial learning rate is 0.006, the batch size is 64, and the number of iterations is 500 rounds.

[0045] Further, in step 3.3, during the training process, the 3-fold cross-validation method is adopted to verify the multi-layer perceptron neural network model in real time, search for the optimal hyperparameter combination, and dynamically adjust the hyperparameters of the multi-layer perceptron neural network model based on the optimal hyperparameter combination.

[0046] The advantages of the present invention are as follows:

[0047] 1. The design method of the blowing and suction air duct for suppressing transonic shock buffet proposed by the present invention introduces the CST parameterization method, and accurately describes the complex duct shape with the minimum design variables during the CST parameterization process. Then, through the deep neural network surrogate model and the particle swarm optimization (PSO) algorithm, the external shape parameters of the blowing and suction air duct are efficiently optimized, so that the designed blowing and suction air duct meets the airfoil aerodynamic parameter requirements, achieving the purpose of suppressing shock buffet and improving the airfoil aerodynamic performance in the transonic state, and the optimization process is simple.

[0048] 2. In the present invention, by combining the CST parameterization method with the deep neural network surrogate model, a database including the structural parameters of the blowing and suction air duct and the corresponding airfoil aerodynamic parameters is obtained, and the time-consuming of a single optimization is shortened from about 3 hours calculated by the traditional CFD method to about 1.2 seconds, greatly improving the efficiency.

[0049] 3. It is verified that the blowing and suction air duct optimized by the method of the present invention shows good control effects on shock buffet under different Mach numbers and angles of attack. Under the state of Ma = 0.73, the optimized blowing and suction air duct increases the lift coefficient of the airfoil by 1.3%, and at the same time reduces the lift coefficient amplitude to the 10 -4 order of magnitude, and delays the buffet onset angle by 1.5°.

[0050] The additional aspects and advantages of the present invention will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present invention. Description of the Drawings

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0052] Figure 1 This is the flow chart of the design method of the blowing and suction air duct for suppressing transonic shock buffet based on deep learning in the present invention;

[0053] Figure 2 This is the schematic diagram of the multi-layer perceptron neural network model constructed in the present invention;

[0054] Figure 3 This is the comparison chart of the geometric shapes of the blowing and suction air ducts before and after optimization. The left figure is the configuration diagram of the blowing and suction air duct before optimization, and the right figure is the configuration diagram of the blowing and suction air duct optimized by the method of the present invention;

[0055] Figure 4 This is the comparison chart of the pressure nephograms of the airfoil before and after optimization. The left figure is the pressure diagram of the airfoil with the traditional blowing and suction air duct, and the right figure is the pressure diagram of the airfoil with the blowing and suction air duct optimized by the present invention.

[0056] Explanation of reference numerals: 1 - suction port, 2 - blowing port, 3 - left wall surface, 4 - right wall surface. Detailed implementation manners

[0057] The embodiments of the present invention will be described in detail below. The described embodiments are exemplary and are intended to explain the present invention and should not be construed as a limitation to the present invention.

[0058] A blowing and suction air duct includes a suction port 1, a blowing port 2, and a duct provided between the suction port and the blowing port. The duct includes a left wall surface 3 and a right wall surface 4. Based on the optimization design of the structural parameters of this type of blowing and suction air duct, this embodiment refers to the Figure 1 process to provide a design method of the blowing and suction air duct for suppressing transonic shock buffet based on deep learning, which specifically includes the following steps:

[0059] Step 1: Obtain the original coordinates of the duct wall surface, describe the geometric shape of the blowing and suction air duct by using the CST parameterization method, construct the shape functions of the left wall surface 3 and the right wall surface 4, determine the optimization variables of the blowing and suction air duct and perform variable parameter adjustment, and randomly generate multiple configurations of the blowing and suction air duct to obtain the structural parameters of the blowing and suction air duct. Specifically, it includes:

[0060] Sub-step 1.1: Extract the original coordinates (x, y) of the duct wall surface, and perform normalization processing through the following formula to eliminate the influence of dimensions and map all the original coordinates of the duct wall surface to the interval [0, 1].

[0061] The normalization processing formula is:

[0062]

[0063] In the formula: x iis the dimensionless representation of the runner wall coordinate, x is the original coordinate of the runner wall, min(x) is the minimum value of the original coordinate of the runner wall, and max(x) is the maximum value of the original coordinate of the runner wall.

[0064] Sub-step 1.2: Use the CST parameterization method to parametrically describe the left and right walls of the blowing and suction runner, and obtain the parametric description equations of the left and right walls:

[0065] The parametric description equation of the left wall is: Y left (x i ) = C(x i )S left (x i ) + x i Y LE

[0066] The parametric description equation of the right wall is: Y right (x i ) = C(x i )S right (x i ) + x i Y RE

[0067] In the above formula, C(x i ) is the category function, S left (x i ) and S right (x i ) are the left wall shape function and the right wall shape function respectively, and are constructed by Bernstein polynomials; Y LE and Y RE are the x coordinates of the left and right walls after normalization respectively, where the coordinate origin is located at the leading edge of the airfoil, the x-axis is along the chord of the airfoil, and the y-axis is along the span of the airfoil.

[0068] Sub-step 1.3: Construct the shape function, determine the runner optimization variables and the runner geometric constraint conditions.

[0069] When constructing the left wall shape function and the right wall shape function using Bernstein polynomials, in this embodiment, the weight coefficients of the 6th-order sub-shape functions are selected as the design parameters for runner optimization, which not only ensures the degree of freedom of runner shape control but also does not introduce too many design variables. The expression of the Bernstein polynomial S i (x) is:

[0070]

[0071] In the formula, B k,6 (x i ) is the 6th-order Bernstein basis function, and w k is the weight coefficient;

[0072] In this step, 5 component shape functions are respectively selected at corresponding positions on the left and right wall surfaces as variables for the optimization of the flow channel. The geometric shape changes of the flow channel are flexibly described by these shape functions. Moreover, the position coordinates of the suction port and the blowing port of the blowing and suction flow channel are selected as the geometric constraint conditions of the flow channel to ensure that the head and tail coordinates of the left and right wall surfaces remain unchanged, guarantee the accuracy of the superposition of the deformed flow channel and the airfoil surface, that is, the geometric continuity of the airfoil at the suction port and the blowing port of the flow channel, and ensure the smooth transition of the curvature of the airfoil wall surface.

[0073] Sub-step 1.4: Generate a parametric program for the flow channel design based on MATLAB, adjust the variable parameters of the flow channel optimization variables, and randomly generate multiple blowing and suction flow channel configurations based on the LHS method.

[0074] Step 2: Mesh the internal flow field of some of the randomly generated blowing and suction flow channels in Step 1, and perform flow field calculations to obtain the airfoil aerodynamic parameters corresponding to different configurations of the blowing and suction flow channels, and establish a blowing and suction flow channel database including the structural parameters of the blowing and suction flow channels and the corresponding airfoil aerodynamic parameters. The specific process is as follows:

[0075] Sub-step 2.1: Select some of the blowing and suction flow channels with randomly generated configurations in Step 1, and mesh the internal flow field of the blowing and suction flow channels through Pointwise to generate a number of meshes.

[0076] Sub-step 2.2: Import the generated meshes into the Fluent simulation software for flow field calculations to obtain the key airfoil aerodynamic parameters corresponding to each configuration of the blowing and suction flow channels. The key airfoil aerodynamic parameters include the lift coefficient and the drag coefficient.

[0077] Sub-step 2.3: Feed back the airfoil lift coefficient and drag coefficient corresponding to different configurations obtained in Step 2.2, as well as the structural parameters of the blowing and suction flow channels to MATLAB to obtain the blowing and suction flow channel database. In this embodiment, 170 representative flow channel parameter samples are generated by MATLAB for the subsequent training and prediction of the neural network model.

[0078] Step 3: Establish and train a multi-layer perceptron neural network model based on TensorFlow to establish the mapping relationship between the flow channel parameters and the aerodynamic performance.

[0079] Sub-step 3.1: Construct a multi-layer perceptron neural network model based on TensorFlow. The multi-layer perceptron neural network model includes an input layer, a fully connected layer, a batch normalization layer, an activation layer, a Dropout layer, and an output layer.

[0080] Refer to Figure 2Schematic diagram of the multi-layer perceptron neural network model. The multi-layer perceptron (MLP) model is based on the TensorFlow framework and includes an input layer (10-dimensional), two hidden layers (64→32→16 nodes), an output layer (2-dimensional) fully connected layer, a batch normalization layer, and an activation layer. The added Dropout layer is used to prevent overfitting during model training and prediction.

[0081] Set the hyperparameters of the multi-layer perceptron neural network model: In the model compilation stage, the Adam optimizer was selected, the initial learning rate was set to 0.006, the batch size was set to 64, and the number of iterations was 500 rounds. The mean squared error was used as the loss function to quantify the gap between the model's predicted value and the true value. Its expression is as follows:

[0082]

[0083] In the formula, y i is the true value of the sample, y p is the predicted value of the sample, and n is the number of samples.

[0084] Step 3.2: Use the random sampling method to divide the 170 runner parameter samples generated in Step 2 into a training set and a test set according to a ratio of 9:1. Among them, the training set has 153 samples, and the test set has 17 samples. The training set is used for subsequent neural network learning and parameter optimization, while the test set is used to evaluate the generalization ability of the model.

[0085] Step 3.3: Use the training set data in Step 3.2 to train the multi-layer perceptron neural network model, and use the 3-fold cross-validation method to perform real-time verification on the multi-layer perceptron neural network model during the training process. Determine the optimal hyperparameter combination through grid search, and dynamically adjust the hyperparameters of the multi-layer perceptron neural network model according to the obtained optimal hyperparameter combination to adjust the model training strategy and ensure that the prediction error of the test set is less than 5%. During the entire training stage, the model goes through 500 training cycles, and 64 samples are randomly selected from the determined training set samples for training in each batch.

[0086] Step 4: Input the remaining blowing and suction runner structure parameters in Step 1 into the multi-layer perceptron neural network model trained and predicted in Step 3, and output the corresponding mean value and amplitude of the airfoil lift coefficient.

[0087] Step 5: Based on the blowing and suction runner database established in Step 2, as well as the blowing and suction runner structure parameters and the corresponding mean value and amplitude of the airfoil lift coefficient obtained based on the multi-layer perceptron neural network model in Step 4, construct an optimized database for the blowing and suction runner structure parameters.

[0088] Step 6: First, establish an objective optimization function with the average value of the maximum lift coefficient and the fluctuation amplitude of the minimum lift coefficient of the blowing and suction air flow channels as the optimization objectives. The objective optimization function is as follows:

[0089] minf val =-C L +ΔC L

[0090] In the formula, f val is the objective function value, and minf val represents the comprehensive performance index to be minimized. C L is the average value of the normalized lift coefficient, and ΔC L is the fluctuation amplitude of the normalized lift coefficient.

[0091] Then, based on the pyswarms library, run the PSO algorithm. Use the PSO algorithm to perform global optimization in the optimization database of the blowing and suction air flow channel structure parameters established in Step 5, and perform double-objective optimization calculation on the average value of the maximum lift coefficient and the fluctuation amplitude of the minimum lift coefficient to obtain the optimal blowing and suction air flow channel structure parameters and the corresponding mean value and amplitude of the airfoil lift coefficient.

[0092] In this step, the PSO optimization algorithm is adopted. By dynamically adjusting the acceleration coefficient and the inertia weight for global optimization, the global search ability and the local search ability are better balanced, and the robustness and adaptability of the algorithm are improved.

[0093] Figure 3 is the comparison diagram of the geometric shapes of the blowing and suction air flow channels before and after optimization. Figure 4 is the comparison of the corresponding airfoil pressure nephograms before and after optimizing the blowing and suction air flow channels. It can be seen from the figure that the optimized blowing and suction air flow channel shows good control effect on shock buffet under different Mach numbers and angles of attack. Especially in the state of Ma = 0.73, the optimized blowing and suction air flow channel increases the lift coefficient of the airfoil by 1.3%, and at the same time reduces the amplitude of the lift coefficient to the order of 10 -4 , and the buffet onset angle is postponed by 1.5°, improving the aerodynamic performance of the airfoil.

[0094] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention.

Claims

1. A design method of a blowing and suction air duct for suppressing transonic shock buffet based on deep learning. The blowing and suction air duct includes a suction port, a blowing port, and a duct disposed between the suction port and the blowing port. The duct includes a left wall surface and a right wall surface, and is characterized in that The described design method includes the following steps: Step 1: Obtain the original coordinates of the runner wall surface, describe the geometry of the blowing and suction runner using the CST parameterization method, construct the shape functions of the left wall surface and the right wall surface, determine the optimization variables of the blowing and suction runner and perform variable parameter adjustment, randomly generate multiple configurations of the blowing and suction runner, and obtain the structural parameters of the blowing and suction runner; Step 2: Mesh the internal flow field of some of the blowing and suction runners randomly generated in Step 1, and perform flow field calculations to obtain the airfoil aerodynamic parameters corresponding to different configurations of the blowing and suction runner, and establish a blowing and suction runner database including the structural parameters of the blowing and suction runner and the corresponding airfoil aerodynamic parameters; Step 3: Construct a multi-layer perceptron neural network model based on TensorFlow and perform training and prediction; the input of the multi-layer perceptron neural network model is the structural parameters of the blowing and suction runner, and the output is the mean value and amplitude of the airfoil lift coefficient; Step 4: Input the remaining structural parameters of the blowing and suction runner in Step 1 into the multi-layer perceptron neural network model trained and predicted in Step 3, and output the corresponding mean value and amplitude of the airfoil lift coefficient; Step 5: According to the blowing and suction runner database established in Step 2, and the structural parameters of the blowing and suction runner and the corresponding mean value and amplitude of the airfoil lift coefficient obtained based on the multi-layer perceptron neural network model in Step 4, construct an optimization database for the structural parameters of the blowing and suction runner; Step 6: Establish an objective optimization function with the average value of the maximum lift coefficient and the fluctuation amplitude of the minimum lift coefficient of the blowing and suction runner as the optimization objectives, perform global optimization using the PSO optimization algorithm in the optimization database for the structural parameters of the blowing and suction runner constructed in Step 5, and solve to obtain the optimal structural parameters of the blowing and suction runner and the corresponding mean value and amplitude of the airfoil lift coefficient; The objective optimization function is: min f val = -C L + ΔC L where: f val is the objective function value, min f val represents the comprehensive performance index to be minimized, C L is the average value of the normalized lift coefficient, ΔC L is the fluctuation amplitude of the normalized lift coefficient.

2. The design method according to claim 1, characterized in that The said Step 1 includes the following sub-steps: Step 1.1: Obtain the original coordinates (x, y) of the left wall surface and the right wall surface of the blowing and suction runner, and perform normalization processing to map the original coordinates of the wall surface of the blowing and suction runner to the interval [0, 1]; The normalization processing formula is: where x i is the dimensionless representation of the runner wall coordinate, x is the original coordinate of the runner wall, min(x) is the minimum value of the original coordinate of the runner wall, and max(x) is the maximum value of the original coordinate of the runner wall; Step 1.2: Use the CST parameterization method to establish the parameterization description equations of the left wall surface and the right wall surface, The parameterization description equation of the left wall surface is: Y left (x i )=C(x i )S left (x i )+x i Y LE The parameterization description equation of the right wall surface is: Y right (x i ) = C(x i )S right (x i ) + x i Y RE Wherein, C(x i ) is a category function, and S left (x i ) and S right (x i ) are the left wall shape function and the right wall shape function respectively, which are constructed by Bernstein polynomials; Y LE and Y RE are the x - coordinates of the left wall and the right wall after normalization respectively, where the coordinate origin is located at the leading edge of the airfoil, the x - axis is along the chord direction of the airfoil, and the y - axis is along the longitudinal direction of the airfoil; Step 1.3: Construct the shape function, determine the runner optimization variables and the runner geometric constraint conditions; Construct shape functions: Select the weight coefficients of the 6th-order sub-shape functions as the runner design parameters, and use the Bernstein polynomials to construct the left wall shape function and the right wall shape function; the expression of the Bernstein polynomial S i (x) is as follows: where B k,6 (x i ) is a 6th-order Bernstein basis function, and w k is the weight coefficient; Select the position coordinates of the blowing port and the suction port of the blowing and suction runner as the runner geometric constraint conditions; And select several component shape functions as the runner optimization variables; Step 1.4: Generate a runner design parameterization program, perform variable parameter adjustment on the runner optimization variables, and randomly generate multiple configurations of the blowing and suction runner.

3. The design method according to claim 2, characterized in that, In the said Step 1.3, 5 component shape functions are respectively selected at the corresponding positions on the left wall surface and the right wall surface as the runner parameter optimization variables.

4. The design method according to claim 3, characterized in that In the said Step 1.4, based on MATLAB, generate a runner design parameterization program, perform variable parameter adjustment on the runner optimization variables; and randomly generate multiple configurations of the blowing and suction runner based on the LHS method.

5. The design method according to claim 1, characterized in that The said Step 2 includes the following sub-steps: Step 2.1: Based on Pointwise software, perform mesh generation on the internal flow field of the partially randomly generated blowing and suction air channels in Step 1 to generate a number of meshes; Step 2.2: Import the generated meshes into the Fluent simulation software for flow field calculation to obtain the airfoil aerodynamic parameters corresponding to different configurations of the blowing and suction air channels, where the airfoil aerodynamic parameters include lift coefficient and drag coefficient; Step 2.3: Establish a blowing and suction air channel database based on MATLAB, where the blowing and suction air channel database includes the structural parameters of different configurations of blowing and suction air channels, as well as the corresponding airfoil lift coefficient and drag coefficient.

6. The design method according to claim 1, wherein The said Step 3 includes the following sub-steps: Step 3.1: Construct a multi-layer perceptron neural network model based on TensorFlow, where the multi-layer perceptron neural network model includes an input layer, a fully connected layer, a batch normalization layer, an activation layer, a Dropout layer, and an output layer; Set the hyperparameters of the multi-layer perceptron neural network model; Step 3.2: Divide the dataset included in the blowing and suction air channel database established in Step 2 into a training set and a test set; Step 3.3: Use the data in the training set to train the multi-layer perceptron neural network model, and perform real-time verification on the model during the training process, and dynamically adjust the hyperparameters of the multi-layer perceptron neural network model; Step 3.4: Use the test set to predict the trained multi-layer perceptron neural network model.

7. The design method according to claim 6, wherein In the said Step 3.1, the multi-layer perceptron neural network model adopts the Adam optimizer; the hyperparameters include the initial learning rate, batch size, and number of epochs, and the initial learning rate is 0.006, the batch size is 64, and the number of epochs is 500 rounds.

8. The design method according to claim 7, characterized in that In the said Step 3.3, during the training process, the 3-fold cross-validation method is used to perform real-time verification on the multi-layer perceptron neural network model, search for the optimal hyperparameter combination, and dynamically adjust the hyperparameters of the multi-layer perceptron neural network model based on the optimal hyperparameter combination.

Images