Nacelle profile optimization method and system based on CST parameterization and intelligent algorithm, and storage medium

Abstract

The invention discloses a nacelle molded line optimization method and system based on CST parameterization and an intelligent algorithm and a storage medium, and the method comprises the steps: carrying out the parameterization modeling of an external molded line of a nacelle through employing a shape-like function method (CST), and taking a weight factor of a Bernstein polynomial as a design variable; setting a value range of a weight factor, and generating a test design matrix through a Latin hypercube sampling method; performing flow field simulation on the nacelle molded line corresponding to each group of weight factors based on ANSYS FLUENT, and calculating a resistance coefficient; constructing a back propagation neural network model, and performing training by taking the weight factor as input and the resistance coefficient as output; predicting an extended database by using the trained neural network, and optimizing by using a genetic algorithm to obtain an optimal weight factor combination; and generating a nacelle external molded line according to the optimal weight factor. According to the invention, through the weight factor based on the linear fitting polynomial, the optimization space is ensured under the condition that the optimization dimension is less, so that the nacelle scheme design efficiency is greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of aircraft aerodynamic shape optimization technology, and particularly relates to a nacelle line optimization method, system and storage medium based on CST parameterization and intelligent algorithm. Background Technology

[0002] Nacelle line optimization refers to the refined design of the geometry of an aircraft engine nacelle (i.e., the outer casing that encloses the engine) to reduce aerodynamic drag, improve airflow efficiency, and enhance overall performance. Existing nacelle line optimization schemes are all based on geometric model parameters, such as setting the leading edge radius, maximum radius, length at the maximum radius, and angle between the trailing edge and the axis of the nacelle line as optimization variables. However, to reduce dimensionality, the nacelle line length, trailing edge radius, and leading edge radius of the lip are set to constant values, thus reducing the optimization space. Summary of the Invention

[0003] (a) Technical problems to be solved

[0004] To address the shortcomings of existing technologies, this invention, by using weighting factors based on linear fitting polynomials, ensures optimization space with fewer optimization dimensions, thereby significantly improving the efficiency of nacelle design.

[0005] (II) Technical Solution

[0006] To achieve the above objectives, the present invention provides the following technical solution: a stub hull line optimization method based on CST parameterization and intelligent algorithms, comprising:

[0007] Step S1: Parametrically model the external profile of the nacelle using the Common Shape Function (CST) method, with the weighting factors of the Bernstein polynomial as the design variables;

[0008] Step S2: Set the range of values ​​for the weighting factors and generate the experimental design matrix using the Latin hypercube sampling method;

[0009] Step S3: Perform flow field simulation on the nacelle profile corresponding to each set of weighting factors based on ANSYS FLUENT, and calculate the drag coefficient;

[0010] Step S4: Construct a backpropagation neural network model and train it with weight factors as input and resistance coefficients as output;

[0011] Step S5: Expand the database using the trained neural network prediction and use a genetic algorithm to find the optimal combination of weight factors.

[0012] Step S6: Generate the nacelle outer profile based on the optimal weighting factor.

[0013] As a preferred embodiment, step S1 specifically involves: establishing a parametric model of the nacelle's external profile using the shape function analogy (CST) method, wherein:

[0014] Describe the geometric features of the profile using dimensionless coordinate parameters;

[0015] The shape equation is constructed based on the product combination of the category function and the shape function.

[0016] As a preferred approach, the Bernstein polynomial weighting factors are solved using the coefficient matrix A, the weight matrix W, and the curve point matrix B, and the least squares method is used to process the linear equation system.

[0017] As a preferred approach, the maximum fitting error between the parameterized profile and the prototype profile is determined by iteratively adjusting the polynomial order n until the maximum fitting error between the parameterized profile and the prototype profile meets a preset threshold. to The parameterized curve is compared with the prototype curve using the maximum fitting error method, as shown in the following formula:

[0018]

[0019] in, This represents the maximum fitting error. This represents the i-th original value. The i-th fitted value is represented by 100 discrete points. The error between the parameterized coordinate points and the actual profile coordinate points is calculated.

[0020] As a preferred embodiment, the range of values ​​for the weighting factor is defined by any of the following methods:

[0021] a) Add or subtract a fixed value based on the original weight value;

[0022] b) Based on the original weight values, add or subtract percentages;

[0023] c) Generate inner and outer boundary profiles based on the original profile offset, and extract their weight factors as boundaries.

[0024] As a preferred embodiment, in step S2, the number of trials is based on the optimization order n. Next, the Latin hypercube sampling method is used to sample within the threshold to generate the experimental design matrix; in the calculation of the drag coefficient, the maximum value of the radial line connecting the upper and lower nacelle lines is used as the reference area.

[0025] As a preferred embodiment, in step S4, 80% of the experimental data is used to train a three-layer backpropagation neural network model, and the remaining 20% ​​is used to verify the model's accuracy. The trained neural network model is then used in conjunction with LHS sampling to generate M (M=100) new sets of data. to And the corresponding drag coefficient results.

[0026] As a preferred embodiment, in step S5:

[0027] The experimental design matrix was generated using Latin hypercube sampling, and the nacelle profile was modified by changing the weighting factor and the drag coefficient was calculated.

[0028] During sampling, each dimension is divided into equally probable sub-intervals, and random sampling within the sub-intervals is achieved through random permutation and uniform distribution.

[0029] A nacelle hull line optimization system based on CST parameterization and intelligent algorithms includes:

[0030] Parametric Modeling Module: Used to parametrically model the external profile of the nacelle using the CST method, with Bernstein polynomial weighting factors as design variables;

[0031] Experimental matrix generation module: used to set the range of weight factor values ​​and generate an experimental design matrix through Latin hypercube sampling;

[0032] Flow field simulation module: used to perform flow field simulation on the profile corresponding to each set of weight factors and calculate the drag coefficient based on CFD tools;

[0033] Proxy model training module: used to build and train a neural network model with weight factors as input and resistance coefficients as output;

[0034] Optimization module: Used to expand the database using the trained neural network and to find the optimal combination of weight factors through a genetic algorithm;

[0035] Shape generation module: used to generate the external shape of the target nacelle based on the optimal weighting factor.

[0036] A computer-readable storage medium storing computer instructions that, when executed by a processor, implement the slack profile optimization method.

[0037] (III) Beneficial Effects

[0038] Compared with existing technologies, this invention provides a nacelle hull line optimization method, system, and storage medium based on CST parameterization and intelligent algorithms, which has the following beneficial effects:

[0039] This invention employs the Shape-Standard Triangulation (CST) method to construct a parametric model of the nacelle's external profile, using fourth-order Bernstein polynomial weighting factors as design variables. An experimental design matrix is ​​generated through Latin hypercube sampling, and the drag coefficient is calculated using CFD simulation. Next, a backpropagation neural network surrogate model is established to predict the drag coefficient. Finally, a genetic algorithm is used to find the optimal combination of weighting factors. This method reduces the optimization dimension from traditional geometric parameters to polynomial weighting factors, further reducing the drag coefficient compared to the original scheme, significantly improving design efficiency, and is suitable for the aerodynamic shape optimization design of aero-engine nacelles. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the optimized method flow of the present invention;

[0041] Figure 2 This is a schematic diagram comparing the original nacelle outline of this invention with the CST parametric outline;

[0042] Figure 3 This is a schematic diagram of the numerical simulation calculation model of the external flow field of the nacelle according to the present invention;

[0043] Figure 4 This is a schematic diagram of the structured mesh of a two-dimensional nacelle used in the numerical simulation of the external flow field of the nacelle according to the present invention.

[0044] Figure 5 This is a structural diagram of the BP neural network model of the present invention;

[0045] Figure 6 This is a flowchart of the optimization process of this invention;

[0046] Figure 7 This is a flowchart of the Ga algorithm of the present invention;

[0047] Figure 8 This is a comparison diagram of the profile lines of the present invention and the prototype lines. Detailed Implementation

[0048] To better understand the purpose, structure, and function of this invention, the following will further explain, in conjunction with the accompanying drawings and specific embodiments, a nacelle hull line optimization method based on CST parameterization and intelligent algorithms.

[0049] Example 1

[0050] refer to Figure 1 This invention discloses a stub hull line optimization method based on CST parameterization and intelligent algorithms. The method includes:

[0051] Step S1: Parametrically model the external profile of the nacelle using the Common Shape Function (CST) method, with the weighting factors of the Bernstein polynomial as the design variables;

[0052] Step S2: Set the range of values ​​for the weighting factors and generate the experimental design matrix using the Latin hypercube sampling method;

[0053] Step S3: Perform flow field simulation on the nacelle profile corresponding to each set of weighting factors based on ANSYS FLUENT, and calculate the drag coefficient;

[0054] Step S4: Construct a backpropagation neural network model and train it with weight factors as input and resistance coefficients as output;

[0055] Step S5: Expand the database using the trained neural network prediction and use a genetic algorithm to find the optimal combination of weight factors.

[0056] Step S6: Generate the nacelle outer profile based on the optimal weighting factor.

[0057] This embodiment proposes to parametrically model the internal and external hull lines of the nacelle using the Class Shape Transformation (CST) method. This method uses a class function and a shape function to describe the aircraft's geometry, providing a high-precision fitting curve with fewer parameters. The parametric model of the nacelle's external hull line is established using CST, where: dimensionless coordinate parameters describe the geometric features of the hull line; and the hull line equation is constructed based on the product of the class function and the shape function. Specifically, in step S1 of this embodiment, the lower hull line (or upper hull line, or inner hull line, or the entire outer hull line) of the two-dimensional nacelle is parametrically modeled using CST to establish the parametric model of the external hull line. Here, X represents the dimensionless x-coordinate of the nacelle's outer hull line, Y represents the y-coordinate, and represents the y-coordinate of the trailing edge of the outer hull line. The calculation formula for the class shape function is established as follows:

[0058]

[0059] The expression for the category function is as follows:

[0060]

[0061] In the formula, N1 and N2 are the shape characteristic coefficients. and Different values ​​are used to construct different geometric shapes; the category function of the National Advisory Committee for Aeronautics (NACA) series of round-nosed and pointed-tail wing types is used. =0.5, =1.0; for double circular arc airfoils, take... =1.0, =1.0; while when =0.75, When = 0.75, the category function can define a Sears-Haack spiral body; due to the characteristics of the nacelle's outer shape having a smooth leading edge and a sharp trailing edge, therefore, respectively take = 0.75. =0.5, =1.0:

[0062] Shape function The calculation formula is as follows:

[0063]

[0064] in, Indicates the weighting coefficient. Let n represent the i-th term of a Bernstein polynomial, where n is the order of the polynomial. These are the coefficients of the polynomial; once n is determined, then n+1 polynomial coefficients are needed to describe the outline.

[0065] like Figure 2 The diagram shows a comparison between the original nacelle outline and the CST parametric outline of this invention. In step S1 of this embodiment, UG software is used to collect point data on the nacelle outline, and a parametric model is created using the CST method to obtain the overall fitting curve of the external outline. Comparing the parametric curve with the prototype curve, the maximum fitting error of the upper part of the outline does not exceed 3.9%, and that of the lower part does not exceed 7.5%. This result indicates that the CST parametric method can accurately describe the nacelle's external outline. The weighting factor of the 4th-order Bernstein polynomial is used in this comparison. The results are shown in Table 1 below:

[0066] Table 1. Bernstein polynomial weighting factors for nacelle hull lines

[0067] Furthermore, the weighting coefficients of its Bernstein polynomial are obtained through the coefficient matrix. Weight matrix and the point selection matrix of the profile Solve the following:

[0068]

[0069]

[0070] For a system of linear equations, the least squares method is used; therefore, the formula for calculating the weight factor matrix of the Bernstein polynomial is as follows:

[0071] .

[0072] To lay the foundation for training the surrogate model, the polynomial order n is iteratively adjusted so that the maximum fitting error between the parameterized profile and the prototype profile meets a preset threshold, thereby determining... to In the above external profile parameterization model, an initial order n is defined, and the corresponding weighting factor is obtained by substituting it into the shape function. to The value, based on the weighting factor to Substitute the value into the category shape function The parameterized coordinates of all the parameterized curve points of the lower profile of the nacelle are obtained. The parameterized curve is then compared with the prototype curve using the maximum fitting error method, as shown in the following formula:

[0073]

[0074] in, This represents the maximum fitting error. This represents the i-th original value. The i-th fitted value is represented by 100 discrete points (more discrete points can be used if there are more profiles; in this embodiment, the axial coordinates of 100 discrete points are fixed, i.e., the x-coordinate of each point). The error between the parameterized coordinate points and the actual profile coordinate points is calculated, and it is determined whether the error value meets the conditions.

[0075]

[0076] If the conditions are not met, the initial order n is incremented by 1, and the parameterized coordinates are recalculated using the shape function. The error is then calculated again until the error value meets the conditions. At this point, the optimized order n and the optimized weighting factor are considered to have been obtained. to (The maximum optimization order is 10. The higher the order, the higher the computational cost, but the improvement in accuracy is not significant.)

[0077] Example 2

[0078] This embodiment uses ANSYS FLUENT software to numerically simulate the external flow field of the nacelle. A density-coupled solver is selected to solve the Reynolds-averaged Navier-Stokes equations. The turbulence model uses SST k-ω with the energy equation enabled. An ideal gas is chosen as the fluid domain medium, and a two-dimensional non-axisymmetric nacelle model is used for numerical simulation, which allows for rapid acquisition of the flow field structure and evaluation of the nacelle drag. The transverse and longitudinal lengths of the far-field boundary are 10 times the length and width of the nacelle's outer surface, respectively. The calculation model is as follows: Figure 3 As shown. The model was meshed using ANSYS ICEM software, where... Figure 4 This is a schematic diagram of the structured grid of a two-dimensional nacelle.

[0079] It should be noted that, in this embodiment, the above embodiment 1 is further described, and the range of values ​​for the weighting factor is defined by any of the following methods: a) adding or subtracting a fixed value based on the original weight value;

[0080] b) Add or subtract percentages based on the original weight values; c) Generate inner and outer boundary lines based on the original profile offset, and extract their weight factors as boundaries.

[0081] Specifically, the methods for defining the scope include:

[0082] a. Upper and lower limits: Defining the optimization weight factors to A numerical range is formed by adding and subtracting fixed values ​​respectively;

[0083] b. Percentage definition; for optimizing weighting factors to Add or subtract x% to form numerical ranges;

[0084] c. Curve Offset Framing: Based on the original curve, two curves, inner and outer, are offset to both sides. A similar function method and optimization order n are used to parametrically model the inner and outer curves (the coordinate value x in the nacelle axial direction can be fixed), obtaining the weighting factors corresponding to the inner and outer curves. to and to (or just offset to one side) to or to ).

[0085] Specifically, based on the optimization order n, the number of experiments (simulations) is determined as follows: Next, Latin hypercube sampling (LHS) was used to... to Sampling is performed within a threshold to obtain the Design of Experiments (DOE) matrix, and then the sampled data is analyzed... indivual to The values ​​are used to model the profile and perform simulation calculations to obtain the corresponding drag coefficient, where the drag coefficient is... As shown in the following formula:

[0086]

[0087] In the formula, This represents the sum of pressure drag and friction drag experienced by the nacelle. This indicates the density of the freely flowing medium ahead. This represents the velocity of the freely flowing medium ahead; in two-dimensional calculations, The maximum value of the radial line connecting the upper and lower nacelle profiles is used as the reference area.

[0088] In this embodiment, to optimize the selection of the target, a specific example will be used to further illustrate the concept, taking an aircraft engine nacelle as an example. An aircraft engine nacelle needs to have good aerodynamic performance; therefore, the drag experienced by the nacelle should be as small as possible to improve fuel economy. During cruise, the pressure gradient and velocity variation on the outer surface of the nacelle will lead to changes in pressure drag and friction drag.

[0089] Example 3

[0090] This embodiment describes the process of training the model. In step S4, 80% of the experimental data is used to train the three-layer backpropagation neural network model, and the remaining 20% ​​is used to verify the model's accuracy. The trained neural network model is then used in conjunction with the LHS sampling method to generate M (M=100) new sets of data. to And the corresponding drag coefficient results. Specifically: Select In the group of experimental (simulation) results *80% of the groups were used for training the three-layer backpropagation (BP) neural network model, with the remainder... *The results of the 20% group are used to test the predictive accuracy of the trained model. (The purpose of training the model is to obtain new predictions later.) to The results of CD, different groups to The hull lines represent different nacelles, replacing the traditional CFD method to obtain the large amount of CD result data required. Then, the trained neural network model combined with LHS sampling is used to generate M (M=100) new sets of data. to And the corresponding drag coefficient results.

[0091] Furthermore, the training includes an optimization process, specifically including: sampling, and defining the threshold of the Bernstein polynomial weighting factors according to actual needs, as shown in Table 2 below:

[0092] Table 2. Range of Bernstein polynomial weighting factors

[0093] This embodiment uses Matlab software and Latin Hypercube Sampling (LHS) to obtain the Design of Experiments (DOE) matrix, which includes 5 factors and 25 trials, as shown in Table 3 below. Of these, 20 sets of data are used to train the model, and 5 sets are used to test the model's predictive accuracy.

[0094] Table 3 DOE matrix and its value

[0095]

[0096] The optimization process in step S5 includes using LHS to generate the DOE matrix for simulation, dividing each weight factor dimension into equally probable sub-intervals, determining sampling points through random permutation and uniformly distributed random numbers, updating the weight factor population through binary encoding, crossover, and mutation operations, and performing roulette wheel selection based on fitness (resistance coefficient).

[0097] Furthermore, the optimization process also includes generating a DOE matrix for simulation using LHS (Local Hierarchical Structure). This DOE matrix is ​​generated based on Bernstein polynomial weighting factors, rather than the geometric parameters of the nacelle model. This significantly reduces the dimensionality of the experimental group, making the optimization process more efficient. By changing the Bernstein polynomial weighting factors, the nacelle outline is modified, and then the target value is calculated. LHS (Layered Hierarchical Monte Carlo) is an improved multidimensional hierarchical Monte Carlo sampling technique. Compared to traditional Monte Carlo simulation of random sampling, LHS, through improved sampling strategies, can provide a more uniform and wider sampling space coverage with fewer samples, effectively utilizing sampled values ​​to describe the distribution function of random variables, and providing random combinations. Its calculation formula is as follows:

[0098] Let d be the number of independent variables in this embodiment, 5, to represent the weighting factors. to Each dimension interval is divided into N equally probable sub-intervals (in this embodiment, this refers to the number of simulation trials, 25), where i = 1, 2, ..., N represents the sample number, and j = 1, 2, ..., d represents the dimension. For each dimension j:

[0099] First, generate a random permutation. , making Indicates the first The interval number of each sample in the j-th dimension.

[0100] For each sample, the j-th dimension sampling point represents the value of any weight factor corresponding to any sample in the test matrix. The j-th dimension sampling point can be calculated by the following formula:

[0101]

[0102] Where N represents the number of samples in the experiment matrix; i represents the sample number in the experiment matrix; and j represents the weight factor number. These are random numbers obtained from a uniform distribution U(0, 1). This ensures that there is a sample in each interval while also achieving randomness within each sub-interval.

[0103] Example 4

[0104] This embodiment describes a nacelle profile optimization method based on CST parameterization and intelligent algorithms. Specifically, it focuses on the prediction process, which utilizes a trained neural network model combined with LHS sampling to generate M (M=100) new sets of profiles. to And the corresponding drag coefficient results. For more complex numerical simulation models, a reasonable surrogate model can be selected during the optimization design process to establish the relationship between input and response. The surrogate model can interpolate or fit based on the given sample point data to obtain the estimated response values ​​of unknown points outside the sample points, and finally construct a black box function between the objective function and the input variables, thereby improving efficiency and shortening the optimization design cycle.

[0105] Specifically, such as Figure 5 As shown, a typical three-layer backpropagation (BP) neural network structure consists of an input layer, hidden layers, and an output layer. Assuming the input layer consists of n nodes, it corresponds to n input variables. (i = 1, 2, ..., n). Input variables need to be standardized, primarily to transform the data to a uniform scale, thereby improving the training efficiency and predictive performance of the neural network. Input nodes are weighted... and threshold The linear operations between nodes are passed to the hidden layer. The activation values ​​of each node in the hidden layer are calculated as follows:

[0106] +

[0107] In this context, the output value of each node in the hidden layer is a function of the activation value:

[0108]

[0109] For the output layer, multiple nodes are allowed. Taking a single-node output as an example, the final output z of the network would be as follows:

[0110]

[0111] in, The weights between the hidden layer and the output layer. This is the threshold for the output layer.

[0112] Furthermore, through simulation calculations, different profiles can be obtained. As shown in Table 3 above, the training set of the BP neural network model comprised 80% of all test sets and was randomly selected. Test set 0 represents the original nacelle. (Percentage reduction in drag coefficient is also shown.) The calculation formula is as follows:

[0113]

[0114] in, Indicates the experimental group corresponding .

[0115] Example 5

[0116] refer to Figure 6 This invention discloses a nacelle hull shape optimization method based on CST parameterization and intelligent algorithms. It generates the nacelle external hull shape according to the optimal weighting factors and uses LHS to generate the DOE matrix for simulation. The optimization flowchart is shown below. Figure 6 As shown, the nacelle outline is modified by changing the Bernstein polynomial weighting factor, and then the target value is calculated. However, the amount of data for both independent and dependent variables is limited. The database can be expanded by predicting the dependent variable value within the range of independent variables using a BP neural network model. Finally, the GA algorithm is employed, employing methods such as crossover, mutation, and genetics to find the globally optimal solution. Figure 7 As shown, taking one iteration as an example, for each polynomial weight factor, it is converted into binary code. Then, it undergoes two processes: exchanging partial codes of each weight factor pairwise (crossover process) and modifying individual codes (mutation process) to expand the weight factor database. The fitness of all weight factors is predicted, and the same number of weight factor groups as the initial population are selected for the next round. The weight factor group with the highest fitness automatically enters the next round, while the rest enter the next round through a roulette wheel method; the higher the fitness, the larger the area covered by the roulette wheel. All the above optimization calculation processes are implemented using Matlab software.

[0117] Specifically, taking M=100 groups as an example, the population size is 100, the crossover rate is 0.7, and the mutation rate is 0.005. The optimal result is... =-0.1124, =-0.2733, =-0.2754, =-0.1243, =-0.1137, at this time The predicted value is 0.1353. To verify this result, the external profile of the nacelle was plotted based on the weighting factors of the aforementioned polynomial. Through simulation calculations, the drag coefficient of this optimal nacelle profile is 0.1396, with a relative error of 3.08% compared to the predicted value. This shows that the optimized profile significantly improves aerodynamic performance compared to the original profile. The value was -31.3%, which was better than other test groups. The profile optimization is now complete.

[0118] Further analysis of the results involved optimizing the expanded database using the GA algorithm. To verify the results, the external profile of the nacelle was plotted based on the weighting factors of the aforementioned polynomial, as shown below. Figure 8 As shown, a comparison diagram of the profile and the prototype profile is presented.

[0119] Simulation calculations show that the drag coefficient of the optimal nacelle profile is 0.1396, with a relative error of 3.08% compared to the predicted value. This indicates that the optimized profile significantly improves aerodynamic performance compared to the original profile. The value was -31.3%, which was better than other experimental groups.

[0120] Example 6

[0121] This invention also proposes a nacelle hull line optimization system based on CST parameterization and intelligent algorithms, comprising:

[0122] Parametric Modeling Module: Used to parametrically model the external profile of the nacelle using the CST method, with Bernstein polynomial weighting factors as design variables;

[0123] Experimental matrix generation module: used to set the range of weight factor values ​​and generate an experimental design matrix through Latin hypercube sampling;

[0124] Flow field simulation module: used to perform flow field simulation on the profile corresponding to each set of weight factors and calculate the drag coefficient based on CFD tools;

[0125] Proxy model training module: used to build and train a neural network model with weight factors as input and resistance coefficients as output;

[0126] Optimization module: Used to expand the database using the trained neural network and to find the optimal combination of weight factors through a genetic algorithm;

[0127] Shape generation module: used to generate the external shape of the target nacelle based on the optimal weighting factor.

[0128] The parametric modeling module includes a geometry processing unit that uses the CST algorithm to transform geometric profiles into a Bernstein polynomial weight factor space, achieving dimensionality reduction. The experimental matrix generation module includes a random number generator and a parallel computing unit. Its core task is to efficiently generate a set of design points (sample sets) that possess both good randomness and uniform coverage of the entire design space within a given multidimensional design space—the experimental design matrix. It uses LHS sampling to uniformly sample the weight factor threshold space, ensuring design space coverage. The flow field simulation module includes a CFD solver hardware acceleration card that calls a CFD solver (such as the FVM finite volume method) to calculate the Navier-Stokes equations and output the drag coefficient. The surrogate model training module includes a neural network acceleration chip that uses a BP neural network to fit the weight factors to a nonlinear mapping and optimizes the network weights using mean square error backpropagation. The optimization module includes an evolutionary algorithm optimization engine; a genetic algorithm globally optimizes the response surface constructed by the neural network, avoiding local optima through crossover / mutation. The final profile generation module is the core of the CAD graphics processing, inverting the optimal weight factors into physical coordinates Y(X) to drive the CAD system to generate the geometric model.

[0129] The present invention also includes a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the aforementioned nacelle profile optimization method. The storage medium receives computer instructions, which are then executed by the processor to implement the optimization method and generate optimized profiles. The intelligent algorithm in the method is compiled into executable machine code to achieve "software-defined profile optimization," and the parameters such as weight factor thresholds and error thresholds stored in the medium are consistent with those in the method.

[0130] This invention can find the optimal solution in a short time through scheme optimization, and the drag coefficient is further reduced compared with the original solution. By using the weighting factor based on the linear fitting polynomial, it ensures the optimization space with fewer optimization dimensions, which greatly improves the efficiency of nacelle design.

[0131] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A nacelle profile optimization method based on CST parameterization and intelligent algorithm, characterized in that, The method comprises the following steps: S1, parameterizing modeling of the external profile of the nacelle by using a class shape function method (CST), taking weight factors of Bernstein polynomials as design variables; S2, setting a value range of the weight factors, and generating a test design matrix by using a Latin hypercube sampling method; S3, performing flow field simulation on the profile of the nacelle corresponding to each group of weight factors based on ANSYS FLUENT, and calculating a drag coefficient; S4, constructing a back propagation neural network model, taking the weight factors as input and the drag coefficient as output for training; S5, using the trained neural network to predict an expanded database, and using a genetic algorithm to optimize to obtain an optimal weight factor combination; S6, generating the external profile of the nacelle according to the optimal weight factors.

2. The nacelle profile optimization method based on CST parameterization and intelligent algorithm according to claim 1, characterized in that, The step S1 specifically comprises: establishing a parameterized model of the external profile of the nacelle by using the class shape function method (CST), wherein: dimensionless coordinate parameters are used to describe profile geometric characteristics; profile equations are constructed based on the product combination of class functions and shape functions.

3. The nacelle profile optimization method based on CST parameterization and intelligent algorithm according to claim 1, characterized in that, Bernstein polynomial weight factors are solved by using a coefficient matrix A, a weight matrix W and a profile point matrix B, and a least square method is used to process nonlinear equation groups.

4. The nacelle profile optimization method based on CST parameterization and intelligent algorithm of claim 3, wherein , the maximum fitting error between the parameterized profile and the original profile is adjusted by iteratively adjusting the polynomial order n to meet a preset threshold, to determine to The parameterized curve is compared with the original curve, and the comparison is performed by using the maximum fitting error, as shown in the following formula: wherein, represents the maximum fitting error, represents the i-th original value, represents the i-th fitting value, expressed with 100 discrete points, the error between the parameterized coordinate points and the actual profile coordinate points is calculated.

5. The nacelle profile optimization method based on CST parameterization and intelligent algorithm according to claim 4, characterized in that The value range of the weight factors is framed by any one of the following ways: a) adding or subtracting a fixed value based on an original weight value; b) adding or subtracting a percentage based on the original weight value; c) generating inner and outer boundary profiles based on original profile offset, and extracting the weight factors thereof as boundaries.

6. The nacelle profile optimization method based on CST parameterization and intelligent algorithm of claim 5, wherein , the step S2, according to the optimization order n test number is times, using Latin hypercube sampling method in the threshold value sampling to generate test design matrix; In the drag coefficient calculation, the maximum value of the radial connecting line between the upper and lower profiles of the nacelle is taken as a reference area.

7. The nacelle profile optimization method based on CST parameterization and intelligent algorithm of claim 1, wherein , the step S4 uses 80% of the test data to train a three-layer back propagation neural network model, and the remaining 20% is used to verify the model accuracy, and a trained neural network model is used to generate M groups of new to and the corresponding drag coefficient results.

8. The nacelle profile optimization method based on CST parameterization and intelligent algorithm of claim 1, wherein, In the step S5: a Latin hypercube sampling is used to generate a test design matrix, the profile of the nacelle is modified by changing the weight factors, and the drag coefficient is calculated; when sampling, each dimension is divided into equal-probability subintervals, and random sampling in the subintervals is realized by random arrangement and uniform distribution.

9. A CST parameterization and intelligent algorithm based nacelle profile optimization system, characterized in that, The method comprises: a parameterized modeling module: used for parameterizing modeling of the external profile of the nacelle by using the CST method, taking the weight factors of Bernstein polynomials as design variables; a test matrix generation module: used for setting a value range of the weight factors, and generating a test design matrix by using a Latin hypercube sampling method; a flow field simulation module: used for performing flow field simulation on the profile corresponding to each group of weight factors based on a CFD tool, and calculating a drag coefficient; a proxy model training module: used for constructing a neural network model taking the weight factors as input and the drag coefficient as output, and training; an optimization module: used for expanding a database by using the trained neural network, and obtaining an optimal weight factor combination by using a genetic algorithm; a profile generation module: used for generating the external profile of the nacelle according to the optimal weight factors.

10. A computer-readable storage medium, characterized in that, The computer instructions are stored, and when the instructions are executed by a processor, the nacelle profile optimization method in any one of claims 1-8 is realized.

Images