A rapid evaluation method for mixed laminar flow inhalation effect based on machine learning

Abstract

The present invention discloses a method for rapid evaluation of mixed laminar flow air intake effect based on machine learning, and belongs to the technical field of wind tunnel tests. The method solves the problem that the conventional air intake design-related evaluation methods in the field of mixed laminar flow in the prior art lack evaluation of air intake effect; the present invention samples and establishes a mixed laminar flow wing aerodynamic data set as a sample set for the mixed laminar flow air intake design range and flight conditions of civil aircraft, and pre-processes the independent variables and dependent variables therein respectively; the pre-processed independent variables and dependent variables are modeled based on the Gaussian process regression kernel function to obtain a Gaussian process regression model; the Gaussian process regression model is optimized and adjusted based on the Bayesian optimization method, and the optimized Gaussian process regression model is used to rapidly predict the drag reduction effect and transition suppression effect of different air intake design parameters in the mixed laminar flow. The present invention effectively improves the efficiency of air intake effect evaluation, and can be applied to the prediction of mixed laminar flow drag reduction effect and transition suppression effect.

Description

Technical Field

[0001] The present invention relates to a method for quickly evaluating the mixed laminar flow inhalation effect, and in particular to a method for quickly evaluating the mixed laminar flow inhalation effect based on machine learning, and belongs to the technical field of wind tunnel testing. Background Art

[0002] Drag reduction technology has long been a focus of attention in the design of civil aircraft. It is directly related to the economy and environmental protection of aircraft. For large aircraft, friction drag accounts for about 45-50% of the total drag. Hybrid laminar flow technology expands the laminar flow area through airfoil design, air intake and other means to effectively reduce friction drag. It has become one of the important technical paths to achieve the development of "green aviation". However, in the process of air intake design of hybrid laminar flow wings, it is necessary to repeatedly iterate parameters such as air intake intensity, air intake position, and pore distribution. The demand for numerical simulation and wind tunnel testing of the above process is very large and complex. In future applications, a large number of calculations and tests for different models will greatly increase the R&D cost and cycle.

[0003] In recent years, with the development of machine learning technology, some studies have used machine learning for aerodynamic prediction. However, the relevant predictions are mainly based on flight conditions such as Mach number and angle of attack, as well as traditional geometric shape parameters. In the field of mixed laminar flow, for design parameters such as the number of intake sections, intake position, and intake volume, traditional evaluation methods have long cycles, high costs, and lack relevant evaluation methods for intake effects.

[0004] In summary, a hybrid laminar flow inhalation effect evaluation method with high efficiency is needed for inhalation effect evaluation. Summary of the invention

[0005] A brief overview of the present invention is provided below in order to provide a basic understanding of certain aspects of the present invention. It should be understood that this overview is not an exhaustive overview of the present invention. It is not intended to identify key or important parts of the present invention, nor is it intended to limit the scope of the present invention. Its purpose is merely to present certain concepts in a simplified form as a prelude to a more detailed description discussed later.

[0006] In view of this, in order to solve the problem that the traditional mixed laminar flow field inspiration design related evaluation methods in the prior art lack rapid evaluation of the inspiration effect, the present invention provides a mixed laminar flow inspiration effect rapid evaluation method based on machine learning.

[0007] The technical solution is as follows: A method for rapid evaluation of mixed laminar flow inhalation effect based on machine learning, specifically comprising the following steps:

[0008] S1. Based on the natural laminar wing, aiming at the design range and flight conditions of the mixed laminar air intake of civil aircraft, a mixed laminar wing aerodynamic data set is sampled and established as a sample set, and the independent variables and dependent variables in the sample set are defined;

[0009] S2. Preprocess the sample set based on data characteristics, that is, perform maximum-minimum normalization on the independent variable and z-score normalization on the dependent variable to obtain the preprocessed independent and dependent variables;

[0010] S3. Gaussian process regression The kernel function models the preprocessed independent and dependent variables to obtain a Gaussian process regression model;

[0011] S4. Optimize and adjust the parameters of the Gaussian process regression model based on the Bayesian optimization method, and perform a prediction accuracy test on the optimized Gaussian process regression model to obtain a qualified Gaussian process regression model;

[0012] S5. The tuned and qualified Gaussian process regression model is applied to the mixed laminar flow design and aspiration evaluation to quickly predict the drag reduction effect and transition suppression effect of different aspiration design parameters in the mixed laminar flow.

[0013] Furthermore, the step S1 includes the following steps:

[0014] S11. Select different natural laminar airfoils covering the design range of hybrid laminar air intake of civil aircraft airfoils, and sample the wing design parameters, air intake design parameters and flight parameters of the hybrid laminar airfoil based on the Latin hypercube sampling method to obtain a training set;

[0015] S12. Based on the selected natural laminar airfoil, the Monte Carlo sampling method is used to sample the wing design parameters, suction design parameters and flight parameters of the hybrid laminar airfoil, and the sampling range does not include the training set to obtain a validation set;

[0016] S13. Through e N The coupled RANS method is used to perform CFD numerical simulation on the samples in the training set and the validation set, and the CFD numerical simulation results, namely the natural laminar wing transition position and natural laminar wing drag coefficient of each sample, are obtained;

[0017] S14. adding the airfoil geometry control parameters, the transition position of the natural laminar wing and the drag coefficient of the natural laminar wing of each benchmark natural laminar flow to the corresponding flight parameters in the acquisition training set and the validation set, the above parameters together with the wing design parameters and flight parameters of the hybrid laminar flow wing constitute the independent variables, and the CFD numerical simulation results are used as the dependent variables;

[0018] In S11, the number of samples in the training set accounts for 90% of the total number of samples in the mixed laminar wing aerodynamic data set;

[0019] The Latin hypercube sampling method is used to sample the training set as follows:

[0020] ;

[0021] in, is the dth dimension samples, , N is the number of sampling intervals in the dth dimension, is the upper bound of the dth dimension, is the lower bound of the dth dimension, Represents a random number generated in the interval (0,1);

[0022] In S12, the number of samples in the validation set accounts for 10% of the total number of samples in the mixed laminar wing aerodynamic dataset;

[0023] The Monte Carlo sampling method is used to sample the validation set as follows:

[0024] ;

[0025] in, is the sampling vector of the dth dimension, i.e., the sample in the validation set. , A randomly generated sampling vector with the dth dimension in the interval [0,1] Vectors of the same dimension.

[0026] Furthermore, in S2, the maximum-minimum normalization parameter conversion formula is expressed as:

[0027] ;

[0028] in, is the normalized value, is the original value, is the minimum value of the sample points in the current dimension, is the maximum value of the sample points in the current dimension;

[0029] The z-score standardized parameter conversion formula is expressed as:

[0030] ;

[0031] Among them, Z is the standardized value, X is the original value, is the mean of the sample points in the current dimension, is the standard deviation of the sample points in the current dimension.

[0032] Furthermore, in S3, the pre-processed independent variables and dependent variables are input, and the Gaussian process regression is used to The kernel function is used to build a model, and the mapping relationship between the preprocessed independent variables and the dependent variables is established to obtain the Gaussian process regression model, and regression prediction is performed from the independent variables to the dependent variables. When performing regression prediction, a normalization process and an inverse normalization process are added before and after the Gaussian process regression model. The normalization process is to perform maximum-minimum normalization on the input independent variables, and the inverse normalization process is to perform inverse z-score normalization on the dependent variables obtained by regression prediction.

[0033] The modeling process is expressed as:

[0034] ;

[0035] ;

[0036] ;

[0037] in, is a Gaussian function, is the mean function, is the covariance function, is the predicted target mean, is the covariance vector between the predicted target and the training point, is the transpose of the covariance vector between the predicted target and the training point, is the training output value vector, for Kernel function, is the first spatial sample point, is the second space sample point, is the first spatial sample point and the second spatial sample point The Euclidean distance between is the length scale parameter, is the smoothness parameter, is the gamma function, is the second type of modified Bessel function; the inverse z-score normalization process is expressed as:

[0038] ;

[0039] Among them, Y is the value after inverse normalization, Y PRE is the original value of the prediction, is the mean of the sample points used for training in the current dimension, The standard deviation of the sample points used for training in the current dimension.

[0040] Furthermore, in S4, the parameters for optimizing the adjustment are Length scale parameter in the kernel function and smoothness parameter The optimization process uses the evaluation value R2 of the 10th-order cross-validation as the tuning score, and the Bayesian optimization process adopts Acquisition function;

[0041] The acquisition function is expressed as:

[0042] ;

[0043] ;

[0044] in, is the current best observed value, is the mean, is the standard deviation, and , is the cumulative distribution function, is the probability density function, Adjust parameters for the search;

[0045] Evaluation value R 2 It is expressed as:

[0046] ;

[0047] in, For the True value, For the Prediction values, is the mean of all true values.

[0048] Furthermore, in S5, for the qualified Gaussian process regression model, its input variables are the baseline natural laminar wing parameters, the mixed laminar wing suction design parameters and the flight conditions, and its output variables are the mixed laminar wing drag coefficient and the transition position.

[0049] The beneficial effects of the present invention are as follows: In view of the long evaluation cycle and high cost of the air intake design effect in the current hybrid laminar flow design, the present invention samples and establishes a hybrid laminar flow wing aerodynamic data set as a sample set according to the hybrid laminar flow air intake design range and flight conditions of civil aircraft, pre-processes the sample set data, and uses Gaussian process regression to The kernel function established a rapid evaluation model for the intake effect, adjusted the model parameters through Bayesian parameters, optimized the rapid evaluation model for the intake effect, and realized the rapid evaluation of the intake effect under different flight conditions and design parameters, which helped to reduce the design cost of the mixed laminar intake and shortened the R&D cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0051] Figure 1 A flow chart of a method for rapid evaluation of hybrid laminar inhalation effect based on machine learning;

[0052] Figure 2 The present invention is a flow chart of an embodiment of a method for rapid evaluation of mixed laminar inhalation effect based on machine learning. DETAILED DESCRIPTION

[0053] In order to make the technical solutions and advantages of the embodiments of the present invention more clearly understood, the exemplary embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than an exhaustive list of all the embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0054] refer to Figure 1 and Figure 2 The present embodiment is described in detail, a method for rapid evaluation of mixed laminar flow inhalation effect based on machine learning, specifically comprising the following steps:

[0055] S1. Based on the natural laminar wing, aiming at the design range and flight conditions of the mixed laminar air intake of civil aircraft, a mixed laminar wing aerodynamic data set is sampled and established as a sample set, and the independent variables and dependent variables in the sample set are defined;

[0056] S2. Preprocess the sample set based on data characteristics, that is, perform maximum-minimum normalization on the independent variable and z-score normalization on the dependent variable to obtain the preprocessed independent and dependent variables;

[0057] S3. Gaussian process regression The kernel function models the preprocessed independent and dependent variables to obtain a Gaussian process regression model;

[0058] S4. Optimize and adjust the parameters of the Gaussian process regression model based on the Bayesian optimization method, and perform a prediction accuracy test on the optimized Gaussian process regression model to obtain a qualified Gaussian process regression model;

[0059] S5. The tuned and qualified Gaussian process regression model is applied to the mixed laminar flow design and aspiration evaluation to quickly predict the drag reduction effect and transition suppression effect of different aspiration design parameters in the mixed laminar flow.

[0060] Furthermore, the step S1 includes the following steps:

[0061] S11. Select different natural laminar airfoils covering the design range of hybrid laminar air intake of civil aircraft airfoils, and sample the wing design parameters, air intake design parameters and flight parameters of the hybrid laminar airfoil based on the Latin hypercube sampling method to obtain a training set;

[0062] S12. Based on the selected natural laminar airfoil, the Monte Carlo sampling method is used to sample the wing design parameters, suction design parameters and flight parameters of the hybrid laminar airfoil, and the sampling range does not include the training set to obtain a validation set;

[0063] S13. Through e N The coupled RANS method is used to perform CFD numerical simulation on the samples in the training set and the validation set, and the CFD numerical simulation results, namely the natural laminar wing transition position and natural laminar wing drag coefficient of each sample, are obtained;

[0064] S14. adding the airfoil geometry control parameters, the transition position of the natural laminar wing and the drag coefficient of the natural laminar wing of each benchmark natural laminar flow to the corresponding flight parameters in the acquisition training set and the validation set, the above parameters together with the wing design parameters and flight parameters of the hybrid laminar flow wing constitute the independent variables, and the CFD numerical simulation results are used as the dependent variables;

[0065] In S11, the number of samples in the training set accounts for 90% of the total number of samples in the mixed laminar wing aerodynamic data set;

[0066] The Latin hypercube sampling method is used to sample the training set as follows:

[0067] ;

[0068] in, is the dth dimension samples, , N is the number of sampling intervals in the dth dimension, is the upper bound of the dth dimension, is the lower bound of the d-th dimension, Represents a random number generated in the interval (0,1);

[0069] In S12, the number of samples in the validation set accounts for 10% of the total number of samples in the mixed laminar wing aerodynamic dataset;

[0070] The Monte Carlo sampling method is used to sample the validation set as follows:

[0071] ;

[0072] in, is the sampling vector of the dth dimension, i.e., the sample in the validation set. , A randomly generated sampling vector with the dth dimension in the interval [0,1] Vectors of the same dimension;

[0073] Specifically, the baseline natural laminar wing parameters include airfoil geometry control parameters, wing geometry control parameters, natural laminar wing transition position and natural laminar wing drag coefficient; the mixed laminar wing intake design parameters include intake position ratio, intake section number, intake distribution and intake intensity; the flight condition parameters include Mach number, Reynolds number, angle of attack and sideslip angle.

[0074] Furthermore, in S2, the maximum-minimum normalization parameter conversion formula is expressed as:

[0075] ;

[0076] in, is the normalized value, is the original value, is the minimum value of the sample points in the current dimension, is the maximum value of the sample points in the current dimension;

[0077] The z-score standardized parameter conversion formula is expressed as:

[0078] ;

[0079] Among them, Z is the standardized value, X is the original value, is the mean of the sample points in the current dimension, is the standard deviation of the sample points in the current dimension.

[0080] Furthermore, in S3, the pre-processed independent variables and dependent variables are input, and the Gaussian process regression is used to The kernel function is used to build a model, and a mapping relationship between the preprocessed independent variables and the dependent variables is established to obtain a Gaussian process regression model. Regression prediction from the independent variables to the dependent variables is performed. When performing regression prediction, a normalization process and an inverse normalization process are added before and after the Gaussian process regression model. The normalization process is to perform maximum-minimum normalization on the input independent variables, and the inverse normalization process is to perform an inverse z-score normalization operation on the dependent variables obtained by regression prediction.

[0081] The modeling process is expressed as:

[0082] ;

[0083] ;

[0084] ;

[0085] in, is a Gaussian function, is the mean function, is the covariance function, is the predicted target mean, is the covariance vector between the predicted target and the training point, is the transpose of the covariance vector between the predicted target and the training point, is the training output value vector, for Kernel function, is the first spatial sample point, is the second space sample point, is the first spatial sample point and the second spatial sample point The Euclidean distance between is the length scale parameter, is the smoothness parameter, is the gamma function, is the modified Bessel function of the second kind;

[0086] The inverse z-score normalization process is expressed as:

[0087] ;

[0088] Among them, Y is the value after inverse normalization, Y PRE is the original value of the prediction, is the mean of the sample points used for training in the current dimension, The standard deviation of the sample points used for training in the current dimension.

[0089] Furthermore, in S4, the parameters for optimizing the adjustment are Length scale parameter in the kernel function and smoothness parameter The optimization process uses the evaluation value R2 of the 10th-order cross-validation as the tuning score, and the Bayesian optimization process adopts Acquisition function;

[0090] The acquisition function is expressed as:

[0091] ;

[0092] ;

[0093] in, is the current best observed value, is the mean, is the standard deviation, and , is the cumulative distribution function, is the probability density function, Adjust parameters for the search;

[0094] Evaluation value R 2 It is expressed as:

[0095] ;

[0096] in, For the True value, For the Prediction values, is the mean of all true values.

[0097] Further, in said S5, for the tuned qualified Gaussian process regression model, its input variables are the reference natural laminar wing parameters, the mixed laminar wing suction design parameters and the flight conditions, and its output variables are the mixed laminar wing drag coefficient and the transition position;

[0098] Specifically, the air intake evaluation is based on the completion of the natural laminar flow wing design. Before the prediction begins, the control parameters and aerodynamic parameters of the benchmark natural laminar flow wing for air intake design are known quantities, and does not include air intake design situations where the benchmark airfoil is not natural laminar or the aerodynamic parameters are unknown.

[0099] Although the present invention has been described according to a limited number of embodiments, it will be apparent to those skilled in the art, with the benefit of the above description, that other embodiments may be envisioned within the scope of the invention thus described. In addition, it should be noted that the language used in this specification is selected primarily for readability and teaching purposes, rather than for explaining or defining the subject matter of the present invention. Therefore, many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the appended claims. The disclosure of the present invention is illustrative, not restrictive, with respect to the scope of the present invention, which is defined by the appended claims.

Claims

1. A method for rapid evaluation of mixed laminar inhalation effect based on machine learning, characterized in that: The following steps are involved: S1. Based on the natural laminar wing, aiming at the design range and flight conditions of the mixed laminar air intake of civil aircraft, a mixed laminar wing aerodynamic data set is sampled and established as a sample set, and the independent variables and dependent variables in the sample set are defined; S2. Preprocess the sample set based on data characteristics, that is, perform maximum-minimum normalization on the independent variable and z-score normalization on the dependent variable to obtain the preprocessed independent and dependent variables; S3. Gaussian process regression The kernel function models the preprocessed independent and dependent variables to obtain a Gaussian process regression model; S4. Optimize and adjust the parameters of the Gaussian process regression model based on the Bayesian optimization method, and perform a prediction accuracy test on the optimized Gaussian process regression model to obtain a qualified Gaussian process regression model; S5. Apply the tuned and qualified Gaussian process regression model to the mixed laminar flow design and air intake evaluation, and quickly predict the drag reduction effect and transition suppression effect of different air intake design parameters in the mixed laminar flow; In S3, the preprocessed independent variables and dependent variables are input, and the Gaussian process regression is performed based on The kernel function is used to build a model, and the mapping relationship between the preprocessed independent variables and the dependent variables is established to obtain the Gaussian process regression model, and regression prediction is performed from the independent variables to the dependent variables. When performing regression prediction, a normalization process and an inverse normalization process are added before and after the Gaussian process regression model. The normalization process is to perform maximum-minimum normalization on the input independent variables, and the inverse normalization process is to perform inverse z-score normalization on the dependent variables obtained by regression prediction. The modeling process is expressed as: y(x)~gp(m(x),K(x,x′)) Among them, gp is the Gaussian function, m(x) is the mean function, k(x,x′) is the covariance function, is the predicted target mean, k * is the covariance vector between the predicted target and the training point, is the transpose of the covariance vector between the predicted target and the training point, y(x) is the training output value vector, for Kernel function, x i is the first spatial sample point, x j is the sample point in the second space, ||x i -x j || is the first spatial sample point x i and the second space sample point x j The Euclidean distance between them, l is the length scale parameter, ν is the smoothness parameter, Γ is the gamma function, K ν is the modified Bessel function of the second kind; The inverse z-score normalization process is expressed as: Y=Y PRE ×σ TRAIN +μ TRAIN Among them, Y is the value after inverse normalization, Y PRE is the original value of the prediction, μ TRAIN is the mean of the sample points used for training in the current dimension, σ TRAIN The standard deviation of the sample points used for training in the current dimension; In S4, the parameters for optimizing the adjustment are: The length scale parameter l and smoothness parameter ν in the kernel function, the evaluation value R2 of the 10th order cross-check is used as the tuning score in the optimization process, and the EI acquisition function is used in the Bayesian optimization process; The EI acquisition function is expressed as: EI(x)=(μ(x)-f(x + )-ξ)Φ(Z)+σ(x)φ(Z); Among them, f(x + ) is the current best observation value, μ(x) is the mean, σ(x) is the standard deviation, and σ(x)>0, Φ is the cumulative distribution function, φ is the probability density function, and ξ is the search adjustment parameter; Evaluation value R 2 It is expressed as: Among them, y i is the i-th true value, is the i-th predicted value, is the mean of all true values.

2. A method for rapid evaluation of mixed laminar inhalation effect based on machine learning according to claim 1, characterized in that: The S1 comprises the following steps: S11. Select different natural laminar airfoils covering the design range of hybrid laminar air intake of civil aircraft airfoils, and sample the wing design parameters, air intake design parameters and flight parameters of the hybrid laminar airfoil based on the Latin hypercube sampling method to obtain a training set; S12. Based on the selected natural laminar airfoil, the Monte Carlo sampling method is used to sample the wing design parameters, suction design parameters and flight parameters of the hybrid laminar airfoil, and the sampling range does not include the training set to obtain a validation set; S13. Through e N The coupled RANS method is used to perform CFD numerical simulation on the samples in the training set and the validation set, and the CFD numerical simulation results, namely, the natural laminar wing transition position and the natural laminar wing drag coefficient of each sample, are obtained; S14. adding the airfoil geometry control parameters, transition position of the natural laminar wing and drag coefficient of the natural laminar wing of each benchmark natural laminar flow to the corresponding flight parameters in the training set and the validation set, wherein the above parameters together with the wing design parameters and flight parameters of the hybrid laminar flow wing constitute independent variables, and the CFD numerical simulation results are used as dependent variables; In S11, the number of samples in the training set accounts for 90% of the total number of samples in the mixed laminar wing aerodynamic data set; The Latin hypercube sampling method is used to sample the training set as follows: Among them, x d,i is the i+1th sample in the dth dimension, i=0,1,2...N-1, N is the number of sampling intervals in the dth dimension, a d is the upper bound of the dth dimension, b d is the lower bound of the dth dimension, rand(0,1) represents a random number generated in the interval (0,1); In S12, the number of samples in the validation set accounts for 10% of the total number of samples in the mixed laminar wing aerodynamic dataset; The Monte Carlo sampling method is used to sample the validation set as follows: x d =a d +rand d,n ×(b d -a d ); Among them, x d is the sampling vector of the dth dimension, i.e., the sample in the validation set, x d =(x1,x2...x n ), rand d,n A randomly generated sampling vector x with the dth dimension in the interval [0,1] d Vectors of the same dimension.

3. A method for rapid evaluation of mixed laminar inhalation effect based on machine learning according to claim 2, characterized in that: In S2, the maximum-minimum normalization parameter conversion formula is expressed as: Among them, X scaler is the normalized value, X i is the original value, X min is the minimum value of the sample point in the current dimension, X max is the maximum value of the sample points in the current dimension; The z-score standardized parameter conversion formula is expressed as: Among them, Z is the standardized value, X is the original value, μ is the mean of the sample points in the current dimension, and σ is the standard deviation of the sample points in the current dimension.

4. The method for rapid evaluation of mixed laminar inhalation effect based on machine learning according to claim 3 is characterized in that: In S5, for the tuned and qualified Gaussian process regression model, its input variables are the baseline natural laminar wing parameters, the mixed laminar wing suction design parameters and the flight conditions, and its output variables are the mixed laminar wing drag coefficient and the transition position.

Images