CN114861315A - Two-dimensional impeller profile optimization method based on machine learning - Google Patents
Two-dimensional impeller profile optimization method based on machine learning Download PDFInfo
- Publication number
- CN114861315A CN114861315A CN202210527158.9A CN202210527158A CN114861315A CN 114861315 A CN114861315 A CN 114861315A CN 202210527158 A CN202210527158 A CN 202210527158A CN 114861315 A CN114861315 A CN 114861315A
- Authority
- CN
- China
- Prior art keywords
- turbine
- dimensional
- profile
- grid
- flow field
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 47
- 238000005457 optimization Methods 0.000 title claims abstract description 37
- 238000010801 machine learning Methods 0.000 title claims abstract description 13
- 238000013528 artificial neural network Methods 0.000 claims abstract description 22
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 17
- 230000002068 genetic effect Effects 0.000 claims abstract description 16
- 238000012847 principal component analysis method Methods 0.000 claims abstract description 10
- 230000009467 reduction Effects 0.000 claims abstract description 9
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 238000011156 evaluation Methods 0.000 claims abstract description 8
- 238000004088 simulation Methods 0.000 claims description 26
- 238000012549 training Methods 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 16
- 239000012530 fluid Substances 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 230000003068 static effect Effects 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000006386 neutralization reaction Methods 0.000 claims description 3
- 238000009966 trimming Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000013461 design Methods 0.000 abstract description 16
- 238000012545 processing Methods 0.000 abstract description 3
- 238000010276 construction Methods 0.000 abstract 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/17—Mechanical parametric or variational design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Geometry (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Automation & Control Theory (AREA)
- Aviation & Aerospace Engineering (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Structures Of Non-Positive Displacement Pumps (AREA)
Abstract
The invention discloses a two-dimensional impeller profile optimization method based on machine learning, which comprises the following steps: leaf type parameterization, grid division, database construction, data dimension reduction, neural network construction and genetic algorithm optimization. On the basis of the original two-dimensional blade type grid of the turbine, the H-shaped grid is kept unchanged, only the grid in the O-shaped grid topology is replaced, and the total number of the grid points is kept unchanged, so that physical quantity parameters of a specified position in a flow field can be quickly extracted from the two-dimensional blade type of the modified turbine with any modeling parameters, the geometric adaptability is better, and the data processing efficiency is effectively improved. The reconstruction of the two-dimensional blade profile flow field of the turbine is realized by combining a principal component analysis method and an artificial neural network method, the dimensionality is reduced on the premise of ensuring the precision requirement, the calculated amount is compressed, the time consumption is further reduced, the calculation resource is saved, the rapid evaluation and surrounding rapid optimization of the pneumatic performance of the two-dimensional blade profile of the turbine are realized, and a new thought is provided for the efficient design of the two-dimensional blade profile of the high-performance turbine.
Description
Technical Field
The invention relates to the field of pneumatic optimization design of a turbine, in particular to a two-dimensional blade profile optimization method of the turbine based on machine learning.
Background
The gas compressor, the turbine and other turbines are important components of the aircraft engine, and the pneumatic optimization design of the turbine blades has important significance for improving the pneumatic performance of the aircraft engine. However, the aeroengine has severe internal flow conditions and complicated flow characteristics such as multistage and unsteady flow characteristics, so that the pneumatic design process of aeroengine turbine components such as a compressor and a turbine is particularly complicated, and the problems of multistage matching and the like are encountered. The traditional pneumatic design of the compressor is developed based on experiments or numerical simulation technology, repeated iteration is needed through design, evaluation and redesign, when the problem of multi-objective optimization is faced, the repeated iteration cost is huge, and the high-performance requirement of the pneumatic design of the new generation of aero-engine cannot be met.
As a fundamental unit of the modeling of a turbine blade, the two-dimensional blade profile design has a crucial impact on turbine performance. The existing two-dimensional blade profile design of the turbine mostly adopts the idea of the superposition thickness of the camber line, has strong experience dependence, needs to be combined with experimental or numerical results during engineering application, adjusts blade profile design parameters to perform iterative optimization repeatedly, needs a large amount of experiments and calculation consumption, and has high time cost. Aiming at the high aerodynamic performance requirement of the aerodynamic design of a new generation of aeroengine, a rapid and accurate optimization algorithm needs to be developed urgently to realize the efficient design of the two-dimensional blade profile of the high-performance turbine.
With the development of a machine learning method in recent years, the technology is gradually applied to optimization solution in various fields, however, for the problem of two-dimensional impeller profile optimization of a turbine, the traditional genetic algorithm highly depends on flow field data solved by a computational fluid dynamics method to evaluate the impeller profile performance, the time consumption is large, so that the method is harsh in application conditions and small in application range on the two-dimensional impeller profile optimization of the turbine, engineering design requirements are difficult to meet, and further the method cannot be applied to engineering practice; in addition, the traditional proxy model has the problems of large error, prediction distortion and the like by directly constructing the relationship between the two-dimensional blade profile modeling parameters of the turbine and the target parameters.
Disclosure of Invention
Technical problem to be solved
The invention aims to provide a two-dimensional impeller profile optimization method based on machine learning, which realizes reconstruction of a two-dimensional impeller profile flow field by combining a principal component analysis method and an artificial neural network method, quickly evaluates the pneumatic performance of the two-dimensional impeller profile, realizes quick optimization, and overcomes the defects of large time consumption, harsh application conditions, small application range and difficulty in meeting engineering design requirements of the conventional two-dimensional impeller profile optimization method; the problems of large error, prediction distortion and the like of the traditional proxy model method are solved.
(II) technical scheme
In order to solve the technical problem, the invention provides a two-dimensional blade profile optimization method of a turbine based on machine learning, which comprises the following steps:
firstly, parameterizing a leaf profile;
step two, grid division;
step three, constructing a database;
step four, reducing the dimension of the data;
step five, building a neural network;
step six, optimizing a genetic algorithm;
wherein the leaf parameterization comprises:
carrying out parametric fitting on the two-dimensional blade profile of the turbine, wherein the parametric fitting adopts a method of stacking thickness distribution by adopting a camber line model, the camber line and the thickness distribution both adopt a quartic Bezier curve structure, and the front edge and the tail edge of the two-dimensional blade profile of the turbine after the parametric fitting adopt arc shapes for trimming;
the meshing comprises:
carrying out grid division on an original two-dimensional blade type of the turbine by adopting an O-4H type grid topology, wherein the total number of grid points is m;
for the turbine two-dimensional blade profile generated by constructing the database in the third step, on the basis of the original turbine two-dimensional blade profile grid, keeping an H-shaped grid unchanged, only replacing the grid in the O-shaped grid topology, and keeping the total number of the grid points to be m;
③ the database is constructed by:
taking +/-20% of the modeling parameters of the mean camber line in the step one as the upper and lower boundaries of the sample space to generateForming a sample space by using the modified two-dimensional blade profiles of the turbine, and selecting n two-dimensional blade profiles of the turbine from the sample space as a database by using a Latin hypercube distribution method, wherein n is more than or equal to 100; performing numerical simulation on the n turbine two-dimensional blade profiles contained in the database by using a computational fluid dynamics method to obtain a numerical simulation flow field of the turbine two-dimensional blade profiles, wherein the obtained numerical simulation flow field forms a database X ═ A 1 ,A 2 ,…,A n Wherein, A represents a two-dimensional impeller profile numerical simulation example, a subscript n represents an example number, and A ═ a 1 ,a 2 ,…,a m } T Wherein a represents numerical analog data at one grid point, and the subscript m corresponds to the total number of the grid points;
fourthly, the data dimensionality reduction comprises the following steps:
performing neutralization treatment on the database X in the step three by adopting the following formula to obtain a difference field X * ：
Applying the following formula to the difference field X * The covariance matrix C of (a) is solved:
wherein Var (A) is a variance expressed as
Cov (A) is covariance expressed as
Wherein mu is the mean value of m-dimensional data in the sample A;
taking the eigenvector w corresponding to the maximum first k eigenvalues of the covariance matrix C 1 ，w 2 ，…，w k Performing diagonalization on the covariance matrix C, wherein
Selecting a set of eigenvectors w that can reduce covariance to zero and maximize variance 1 ，w 2 ，…，w k Forming a target projection matrix;
reducing the dimension by adopting a principal component analysis method, so that the k modal flow fields after dimension reduction are mutually independent;
reconstructing the numerical simulation flow field of the two-dimensional blade profile of the turbine by adopting the linear combination of the k modal flow fields to obtain a reconstructed flow field of the two-dimensional blade profile of the turbine;
constructing a neural network comprises:
dividing a database built by selecting n modified two-dimensional impeller profiles in the third step into a training set and a testing set according to the proportion of 3:1, taking the two-dimensional impeller profile modeling parameters related to the parametric fitting in the first step as input parameters of a neural network, taking the weights of k modal flow fields obtained in the fourth step as output, training samples of the training set, constructing the relationship between the two-dimensional impeller profile modeling parameters and the weights of the k modal flow fields obtained by decomposing the main component analysis method, and checking by adopting the testing set;
comparing the reconstructed flow field of the two-dimensional blade profile of the turbine in the fourth step with the numerical simulation flow field of the two-dimensional blade profile of the turbine in the third step, comparing performance parameters such as static pressure rise coefficient, total pressure loss coefficient and the like, adopting the following formula as a determination coefficient between prediction data and real data,
in the formula, K O To reconstruct the flow field results, K C In order to numerically simulate the results of the flow field,
for the result of the averaged reconstructed flow field,
simulating the flow field result for the averaged value;
evaluating the accuracy of a reconstructed flow field of the two-dimensional blade profile of the turbine, wherein the value range of the decision coefficient R is 0 to 1;
if the value of the decision coefficient R tends to 0, the judgment precision is low, and the fifth step is repeated until the value of the decision coefficient R tends to 1, so that the neural network meeting the precision is obtained;
sixthly, the genetic algorithm optimization comprises the following steps:
and (4) determining an optimized objective function value, taking the neural network obtained by training in the step five as an evaluation means of a genetic algorithm, setting a decision vector domain as a sample space domain, optimizing to obtain an optimized two-dimensional impeller profile, and performing calculation verification on an optimization result by using a computational fluid mechanics method.
(III) advantageous effects
The invention provides a two-dimensional blade profile optimization method of a turbine based on machine learning, which has the following beneficial effects compared with the prior art:
on the basis of the original two-dimensional blade-shaped grid of the turbine, the H-shaped grid is kept unchanged, only the grid in the O-shaped grid topology is replaced, and the total number of the grid points is kept to be m, so that the physical quantity parameters of the specified position in the flow field can be quickly extracted from the modified two-dimensional blade-shaped turbine with any modeling parameters, the geometric adaptability is better, the data processing efficiency can be effectively improved, and the engineering design requirements are met;
by adopting the principal component analysis method to reduce the dimension of the original data, on the premise of ensuring the precision requirement, the dimension is reduced, the calculated amount is compressed, the time consumption is reduced, the calculation resource is saved, the aims of quickly evaluating and quickly optimizing the two-dimensional blade profile pneumatic performance of the turbine are fulfilled, and meanwhile, the problems of large error, prediction distortion and the like of the traditional proxy model method are solved.
Drawings
FIG. 1 is a flow chart of a two-dimensional profile optimization method for a turbomachine based on machine learning;
FIG. 2 is a schematic diagram of a two-dimensional profile parametric fit for a turbine;
FIG. 3 is a schematic grid diagram of a two-dimensional blade profile of a turbine using an O-4H type grid topology (1 grid is shown every 3 grid points);
FIG. 4 is a comparison graph of a reconstructed flow field of a two-dimensional profile of a turbine obtained by a two-dimensional profile optimization method of a turbine based on machine learning and a pressure distribution of a numerical simulation flow field of the two-dimensional profile of the turbine;
FIG. 5 is a graph comparing an optimized two-dimensional profile of a turbine with an original two-dimensional profile of a turbine obtained by a two-dimensional profile optimization method based on machine learning.
Detailed Description
The following detailed description of the present invention will be made in conjunction with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.
As shown in FIG. 1, the invention provides a two-dimensional profile optimization method for a turbine based on machine learning, which comprises the following steps:
step one, parameterization of a leaf profile:
carrying out parametric fitting on the two-dimensional blade profile of the turbine, wherein the parametric fitting adopts a method of stacking thickness distribution by adopting a camber line model, the camber line and the thickness distribution both adopt a quartic Bezier curve structure, and the front edge and the tail edge of the two-dimensional blade profile of the turbine after the parametric fitting adopt arc shapes for trimming.
Step two, grid division:
carrying out grid division on an original two-dimensional blade type of the turbine by adopting an O-4H type grid topology, wherein the total number of grid points is m;
keeping an H-shaped grid unchanged on the basis of the grid of the original two-dimensional blade profile of the turbine generated in the database constructing process related to the third step, only replacing the grid in the O-shaped grid topology, and keeping the total number of the grid points as m;
the fixed H-shaped grid has higher automation degree when a genetic algorithm is used for calling a target parameter, and the speed of evaluating the leaf type is further improved; meanwhile, the application range of the genetic algorithm is widened.
Step three, constructing a database:
taking +/-20% of the modeling parameter of the mean camber line in the step one as the upper and lower bounds of a sample space, generating modified two-dimensional impeller blade profiles to form the sample space, and selecting n two-dimensional impeller blade profiles from the sample space by adopting a Latin hypercube distribution method as a database, wherein n is more than or equal to 100; performing numerical simulation on the n turbine two-dimensional blade profiles contained in the database by using a computational fluid dynamics method to obtain a numerical simulation flow field of the turbine two-dimensional blade profiles, wherein the obtained numerical simulation flow field forms a database X ═ A 1 ,A 2 ,…,A n Wherein, A represents a two-dimensional impeller profile numerical simulation example, a subscript n represents an example number, and A ═ a 1 ,a 2 ,…,a m } T Where a represents numerical analog data at one grid point and the subscript m corresponds to the total number of the grid points.
Step four, data dimension reduction:
performing neutralization treatment on the database X in the step three by adopting the following formula to obtain a difference field X * ：
Applying the following formula to the difference field X * The covariance matrix C of (a) is solved:
wherein Var (A) is a variance expressed as
Cov (A) is covariance expressed as
Wherein mu is the mean value of m-dimensional data in the sample A;
taking the eigenvector w corresponding to the maximum first k eigenvalues of the covariance matrix C 1 ，w 2 ，…，w k Performing diagonalization on the covariance matrix C, wherein
Selecting a set of eigenvectors w that can reduce covariance to zero and maximize variance 1 ，w 2 ，…，w k Forming a target projection matrix;
reducing the dimension by adopting a principal component analysis method, so that the k modal flow fields after dimension reduction are mutually independent;
and reconstructing the numerical simulation flow field of the two-dimensional blade profile of the turbine by adopting the linear combination of the k modal flow fields to obtain a reconstructed flow field of the two-dimensional blade profile of the turbine.
Step five, building a neural network:
dividing a database built by selecting n modified two-dimensional impeller profiles in the third step into a training set and a testing set according to the proportion of 3:1, taking the two-dimensional impeller profile modeling parameters related to the parametric fitting in the first step as input parameters of a neural network, taking the weights of k modal flow fields obtained in the fourth step as output, training samples of the training set, constructing the relationship between the two-dimensional impeller profile modeling parameters and the weights of the k modal flow fields obtained by decomposing the main component analysis method, and checking by adopting the testing set;
comparing the reconstructed flow field of the two-dimensional blade profile of the turbine in the fourth step with the numerical simulation flow field of the two-dimensional blade profile of the turbine in the third step, comparing performance parameters such as static pressure rise coefficient, total pressure loss coefficient and the like, adopting the following formula as a determination coefficient between prediction data and real data,
in the formula, K O To reconstruct the flow field results, K C In order to numerically simulate the results of the flow field,
for the result of the averaged reconstructed flow field,
simulating the flow field result for the averaged value;
evaluating the accuracy of a reconstructed flow field of the two-dimensional blade profile of the turbine, wherein the value range of the decision coefficient R is 0 to 1;
and if the value of the decision coefficient R tends to 0, judging that the precision is low, and repeating the step five until the value of the decision coefficient R tends to 1, thus obtaining the neural network meeting the precision.
Step six, optimizing a genetic algorithm:
and (4) determining an optimized objective function value, taking the neural network obtained by training in the step five as an evaluation means of a genetic algorithm, setting a decision vector domain as a sample space domain, optimizing to obtain an optimized two-dimensional impeller profile, and performing calculation verification on an optimization result by using a computational fluid mechanics method.
Example (b):
take the two-dimensional blade profile of NACA65-K48 subsonic turbine designed by German space navigation center (DLR) as an example.
The original NACA65-K48 subsonic turbine two-dimensional blade profile was parametrically fitted according to the blade profile parametrization described in step one, using a mean camber line profile stacking thickness profile method, including 17 profile parameters, as shown in fig. 2. Wherein the size and the position of the blade profile are determined by chord length b, a mounting angle beta and a grid distance t; the shape of the middle arc is defined by a geometric inlet angle alpha 1 Geometric exit angle alpha 2 And Bezier control point parameter Cx 2 、Cx 3 、Cx 4 、Cy 3 Determining; thickness distribution is determined by leading edge angle theta 1 Angle of trailing edge theta 2 Small radius of leading edge R 1 Small radius of trailing edge R 2 And Bezier control point parameter Hx 2 、Hx 3 、Hx 4 、Hy 3 And (6) determining. The values for the above modeling parameters for the original NACA65-K48 subsonic turbine two-dimensional blade profile and the turbine two-dimensional blade profile at the upper and lower boundaries of the database are shown in table 1.
And according to the method in the second step, carrying out meshing on the NACA65-K48 subsonic turbine two-dimensional blade profile by adopting an O-4H type grid topology, wherein the total number of grid points is 2.6 thousands, the number of H grids which are kept unchanged is 2.3 thousands, and the number of grids in the O type grid topology which are changed for attaching the turbine two-dimensional blade profile is 0.3 thousands as shown in figure 3.
And constructing a database according to the method in the step three. When a database is constructed, only 6 parameters of the arc shape in the NACA65-K48 subsonic turbine two-dimensional blade shape determination are changed, the thickness distribution is kept unchanged, the upper and lower bound values of the modification parameters are shown in table 1, a sample space is formed by the two-dimensional blade shapes of the modified NACA65-K48 subsonic turbines, and 200 modified NACA65-K48 subsonic turbines are selected from the sample space as the database by adopting a Latin hypercube distribution method. And (2) keeping the H-shaped grid unchanged on the basis of the original two-dimensional blade grid of the NACA65-K48 subsonic turbine on the modified NACA65-K48 subsonic turbine generated by the database, only replacing the grid in the O-shaped grid topology, and keeping the total number of the grid points to be 2.3 ten thousand by adopting the method in the step two, wherein the H grid which is kept unchanged is 2.3 ten thousand, and the grid in the O-shaped grid topology which is changed for attaching the two-dimensional blade grid of the modified NACA65-K48 subsonic turbine is 0.3 thousand. Numerical simulation is carried out on 200 modified NACA65-K48 two-dimensional blade forms of the subsonic turbine contained in a database by using a Fine Turbo module in computational fluid dynamics software NUECA to obtain a numerical simulation flow field of the modified NACA65-K48 two-dimensional blade forms of the subsonic turbine, and the numerical simulation of each modified NACA65-K48 two-dimensional blade form takes about 500 seconds. The obtained numerical simulation flow field composition database X ═ { a ═ a 1 ,A 2 ,…,A 200 }，A＝{a 1 ,a 2 ,…,a 26000 } T 。
TABLE 1 original leaf Profile parameters and database boundary parameters
| Prototype | Modified upper bound | Modified lower bound | |
| Chord length b | 0.040031 | 0.040031 | 0.040031 |
| Setting angle beta | 67.53274 | 67.53274 | 67.53274 |
| Grid pitch t | 0.022 | 0.022 | 0.022 |
| Geometric inlet angle alpha 1 | 24.5 | 29.4 | 19.6 |
| Geometric exit angle alpha 2 | 20 | 24 | 16 |
| Cx 2 | 0.24 | 0.288 | 0.192 |
| Cx 4 | 0.81 | 0.972 | 0.648 |
| Cx 3 | 0.62 | 0.744 | 0.496 |
| Cy 3 | 0.164 | 0.1968 | 0.1312 |
| Leading edge angle theta 1 | 14 | 14 | 14 |
| Trailing edge angle θ 2 | 2 | 2 | 2 |
| Small radius of leading edge R 1 | 0.00321 | 0.00321 | 0.00321 |
| Small radius of trailing edge R 2 | 0.00535 | 0.00535 | 0.00535 |
| Hx 2 | 0.17 | 0.17 | 0.17 |
| Hx 3 | 0.55 | 0.55 | 0.55 |
| Hx 4 | 0.92 | 0.92 | 0.92 |
| Hy 3 | 0.021 | 0.021 | 0.021 |
And D, performing data dimensionality reduction on the numerical simulation flow field in the database constructed in the step three according to the method in the step four.
By using
Neutralizing the database X constructed in the third step to obtain a difference field X * 。
By using
For difference field X * The covariance matrix C of (a) is solved.
The eigenvector w corresponding to the maximum first 10 eigenvalues of the covariance matrix C is taken 1 ，w 2 ，…，w 10 To the covarianceCarrying out diagonalization on the matrix C; selecting a set of eigenvectors w that can reduce covariance to zero and maximize variance 1 ，w 2 ，…，w 10 And forming an object projection matrix. And reducing the dimension by adopting a principal component analysis method, so that the 10 modal flow fields after dimension reduction are mutually independent. And reconstructing the numerical simulation flow field of the two-dimensional blade profile of the NACA65-K48 subsonic turbine by adopting the linear combination of 10 modal flow fields to obtain the reconstructed flow field of the two-dimensional blade profile of the NACA65-K48 subsonic turbine.
And building a neural network according to the method in the fifth step. And (3) dividing a database built by 200 modified NACA65-K48 subsonic turbine two-dimensional blade profiles selected in the third step into a training set and a test set according to the proportion of 3:1, wherein the training set comprises 150 modified NACA65-K48 subsonic turbine two-dimensional blade profiles, and the test set comprises 50 modified NACA65-K48 subsonic turbine two-dimensional blade profiles. And (3) taking the 17 NACA65-K48 subsonic turbine two-dimensional blade profile modeling parameters related to parameterized fitting in the step one as input parameters of a neural network, taking the weights of the 10 modal flow fields obtained in the step four as output, training samples of a training set, constructing a relation between the NACA65-K48 subsonic turbine two-dimensional blade profile modeling parameters and the weights of the 10 modal flow fields obtained by decomposition of a principal component analysis method, and checking by adopting a test set. Through inspection, the determination coefficients of the static pressure ratio between the front 1-fold chord length of the two-dimensional blade profile of the NACA65-K48 subsonic turbine and the axial chord length 30% of the rear end of the two-dimensional blade profile of the tail edge are about 0.99, the determination coefficient of the total pressure recovery coefficient is 0.841, and a neural network meeting the precision can be obtained.
And (5) carrying out genetic algorithm optimization according to the method in the sixth step. And (4) optimizing by taking the static pressure ratio as an optimization objective function, taking the neural network obtained by training in the fifth step as an evaluation means of a genetic algorithm, and setting a decision vector domain as a sample space domain. In the genetic algorithm, the population scale of each generation is 50, the mutation probability is set to be 0.4, the cross probability is set to be 0.9, the optimization of 200 generations is carried out, 10k times of leaf profile evaluation is accumulated, the total time consumption is about 107s, and the average time consumption of leaf profile evaluation in each time is only 0.01 s. On the same host, the FINE in the commercial software NUMCA is used TM Turbo mouldThe time for performing the leaf profile calculation once is about 500s, and the speed improvement brought by the PCA-ANN flow field reconstruction method provided by the invention can reach 10 5 Magnitude.
A comparison of the optimized NACA65-K48 subsonic turbine two-dimensional blade profile optimized by genetic algorithm with the original NACA65-K48 subsonic turbine two-dimensional blade profile is shown in FIG. 5. The optimized NACA65-K48 two-dimensional blade profile of the subsonic turbine has a static pressure ratio of 1.368, which is 14.7% higher than that of the original NACA65-K48 two-dimensional blade profile of the subsonic turbine. The optimization result is calculated and verified by using a computational fluid mechanics method, and the result shows that the real static pressure ratio is 1.367, the prediction error is less than 0.1%, and therefore the optimization result can be considered to be reliable.
In summary, the embodiment of the invention combines a principal component analysis method and an artificial neural network method to realize the reconstruction of the two-dimensional blade profile flow field of the turbine, quickly evaluate the pneumatic performance of the two-dimensional blade profile of the turbine and realize quick optimization, and compared with the existing two-dimensional blade profile optimization method of the turbine, the method has better geometric adaptability, can effectively improve the data processing efficiency and meet the engineering design requirements; by adopting the principal component analysis method to reduce the dimension of the original data, on the premise of ensuring the precision requirement, the dimension is reduced, the calculated amount is compressed, the time consumption is reduced, the calculation resource is saved, the rapid assessment and rapid optimization of the pneumatic performance of the two-dimensional blade profile of the turbine can be better realized, meanwhile, the problems of large error, prediction distortion and the like of the traditional proxy model method are overcome, and a new thought can be provided for the efficient design of the two-dimensional blade profile of the high-performance turbine.
Claims (1)
1. A two-dimensional impeller profile optimization method based on machine learning is characterized by comprising the following steps:
firstly, parameterizing a leaf profile;
step two, grid division;
step three, constructing a database;
step four, reducing the dimension of the data;
step five, building a neural network;
step six, optimizing a genetic algorithm;
the leaf profile parameterization includes:
carrying out parametric fitting on the two-dimensional blade profile of the turbine, wherein the parametric fitting adopts a method of stacking a camber line model on thickness distribution, the camber line and the thickness distribution adopt a quartic Bezier curve structure, and the front edge and the tail edge of the two-dimensional blade profile of the turbine after parametric fitting adopt a circular arc shape for trimming;
the meshing comprises:
carrying out grid division on an original two-dimensional blade type of the turbine by adopting an O-4H type grid topology, wherein the total number of grid points is m;
for the turbine two-dimensional blade profile generated by constructing the database in the third step, on the basis of the original turbine two-dimensional blade profile grid, keeping an H-shaped grid unchanged, only replacing the grid in the O-shaped grid topology, and keeping the total number of the grid points to be m;
③ the database is constructed by:
taking +/-20% of the modeling parameter of the mean camber line in the step one as the upper and lower bounds of a sample space, generating modified two-dimensional impeller blade profiles to form the sample space, and selecting n two-dimensional impeller blade profiles from the sample space by adopting a Latin hypercube distribution method as a database, wherein n is more than or equal to 100; performing numerical simulation on the n turbine two-dimensional blade profiles contained in the database by using a computational fluid dynamics method to obtain a numerical simulation flow field of the turbine two-dimensional blade profiles, wherein the obtained numerical simulation flow field forms a database X ═ A 1 ,A 2 ,…,A n Wherein, A represents a two-dimensional impeller profile numerical simulation example, a subscript n represents an example number, and A ═ a 1 ,a 2 ,…,a m } T Wherein a represents numerical analog data at one grid point, and the subscript m corresponds to the total number of the grid points;
fourthly, the data dimensionality reduction comprises the following steps:
performing neutralization treatment on the database X in the step three by adopting the following formula to obtain a difference field X * ：
Applying the following formula to the difference field X * The covariance matrix C of (a) is solved:
wherein Var (A) is a variance expressed as
Cov (A) is covariance expressed as
Wherein mu is the mean value of m-dimensional data in the sample A;
taking the eigenvector w corresponding to the maximum first k eigenvalues of the covariance matrix C 1 ，w 2 ，…，w k Performing diagonalization on the covariance matrix C, wherein
Selecting a set of eigenvectors w that can reduce covariance to zero and maximize variance 1 ，w 2 ，…，w k Forming a target projection matrix;
reducing the dimension by adopting a principal component analysis method, so that the k modal flow fields after dimension reduction are mutually independent;
reconstructing the numerical simulation flow field of the two-dimensional blade profile of the turbine by adopting the linear combination of the k modal flow fields to obtain a reconstructed flow field of the two-dimensional blade profile of the turbine;
constructing a neural network comprises:
dividing a database built by selecting n modified two-dimensional impeller profiles in the third step into a training set and a testing set according to the proportion of 3:1, taking the two-dimensional impeller profile modeling parameters related to the parametric fitting in the first step as input parameters of a neural network, taking the weights of k modal flow fields obtained in the fourth step as output, training samples of the training set, constructing the relationship between the two-dimensional impeller profile modeling parameters and the weights of the k modal flow fields obtained by decomposing the main component analysis method, and checking by adopting the testing set;
comparing the reconstructed flow field of the two-dimensional blade profile of the turbine in the fourth step with the numerical simulation flow field of the two-dimensional blade profile of the turbine in the third step, comparing performance parameters such as static pressure rise coefficient, total pressure loss coefficient and the like, adopting the following formula as a determination coefficient between prediction data and real data,
in the formula, K O To reconstruct the flow field results, K C In order to numerically simulate the results of the flow field,
for the result of the averaged reconstructed flow field,
simulating the flow field result for the averaged value;
evaluating the accuracy of a reconstructed flow field of the two-dimensional blade profile of the turbine, wherein the value range of the decision coefficient R is 0 to 1;
if the value of the decision coefficient R tends to 0, the judgment precision is low, and the fifth step is repeated until the value of the decision coefficient R tends to 1, so that the neural network meeting the precision is obtained;
sixthly, the genetic algorithm optimization comprises the following steps:
and (4) determining an optimized objective function value, taking the neural network obtained by training in the step five as an evaluation means of a genetic algorithm, setting a decision vector domain as a sample space domain, optimizing to obtain an optimized two-dimensional impeller profile, and performing calculation verification on an optimization result by using a computational fluid mechanics method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210527158.9A CN114861315B (en) | 2022-05-16 | Turbine two-dimensional blade profile optimization method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210527158.9A CN114861315B (en) | 2022-05-16 | Turbine two-dimensional blade profile optimization method based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114861315A true CN114861315A (en) | 2022-08-05 |
| CN114861315B CN114861315B (en) | 2024-07-05 |
Family
ID=
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117235911A (en) * | 2023-07-05 | 2023-12-15 | 哈尔滨工业大学 | Two-dimensional modeling method for micro turbine blade of breathing machine |
| CN117313237A (en) * | 2023-09-23 | 2023-12-29 | 哈尔滨工业大学 | Special unmanned aerial vehicle configuration scheme optimization method based on machine learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130118736A1 (en) * | 2010-07-29 | 2013-05-16 | Adam Usadi | Methods and Systems For Machine - Learning Based Simulation of Flow |
| CN105257590A (en) * | 2015-10-20 | 2016-01-20 | 南京航空航天大学 | Semi-serial blade and design method thereof |
| CN108563872A (en) * | 2018-04-16 | 2018-09-21 | 西北工业大学 | Mesh parameterization method and axial flow turbine Aerodynamic optimization design method based on the mesh parameterization method |
| CN111898212A (en) * | 2020-06-17 | 2020-11-06 | 西安交通大学 | Impeller mechanical profile design optimization method based on BezierGAN and Bayesian optimization |
| CN112287580A (en) * | 2020-10-27 | 2021-01-29 | 中国船舶重工集团公司第七0三研究所 | Axial flow compressor surge boundary calculation method based on full three-dimensional numerical simulation |
| CN114154275A (en) * | 2022-02-10 | 2022-03-08 | 北京航空航天大学 | Low-pressure turbine blade profile pneumatic design method based on optimal load distribution model optimization |
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130118736A1 (en) * | 2010-07-29 | 2013-05-16 | Adam Usadi | Methods and Systems For Machine - Learning Based Simulation of Flow |
| CN105257590A (en) * | 2015-10-20 | 2016-01-20 | 南京航空航天大学 | Semi-serial blade and design method thereof |
| CN108563872A (en) * | 2018-04-16 | 2018-09-21 | 西北工业大学 | Mesh parameterization method and axial flow turbine Aerodynamic optimization design method based on the mesh parameterization method |
| CN111898212A (en) * | 2020-06-17 | 2020-11-06 | 西安交通大学 | Impeller mechanical profile design optimization method based on BezierGAN and Bayesian optimization |
| CN112287580A (en) * | 2020-10-27 | 2021-01-29 | 中国船舶重工集团公司第七0三研究所 | Axial flow compressor surge boundary calculation method based on full three-dimensional numerical simulation |
| CN114154275A (en) * | 2022-02-10 | 2022-03-08 | 北京航空航天大学 | Low-pressure turbine blade profile pneumatic design method based on optimal load distribution model optimization |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117235911A (en) * | 2023-07-05 | 2023-12-15 | 哈尔滨工业大学 | Two-dimensional modeling method for micro turbine blade of breathing machine |
| CN117313237A (en) * | 2023-09-23 | 2023-12-29 | 哈尔滨工业大学 | Special unmanned aerial vehicle configuration scheme optimization method based on machine learning |
| CN117313237B (en) * | 2023-09-23 | 2024-04-16 | 哈尔滨工业大学 | Special unmanned aerial vehicle configuration scheme optimization method based on machine learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US7191161B1 (en) | Method for constructing composite response surfaces by combining neural networks with polynominal interpolation or estimation techniques | |
| CN107391891B (en) | Large-aspect-ratio wing optimization design method based on model fusion method | |
| Giannakoglou et al. | Low-cost genetic optimization based on inexact pre-evaluations and the sensitivity analysis of design parameters | |
| US6606612B1 (en) | Method for constructing composite response surfaces by combining neural networks with other interpolation or estimation techniques | |
| CN104697462A (en) | Method for extracting profiling feature parameters of aeronautical blade based on central axis | |
| CN106934074B (en) | Global optimal turbofan engine air inlet channel noise reduction design method | |
| CN111027148B (en) | Automatic calibration and industrial axial flow compressor performance calculation method for loss lag angle model | |
| CN114154275B (en) | Low-pressure turbine blade profile pneumatic design method based on optimal load distribution model optimization | |
| Heidebrecht et al. | Parametric geometry and CFD process for turbofan nacelles | |
| CN112685968A (en) | Axial flow compressor pneumatic design method based on space load customization thought | |
| CN113673027A (en) | Hypersonic aircraft aerodynamic load optimization design method based on proxy model | |
| CN111079326B (en) | Two-dimensional anisotropic grid cell measurement tensor field smoothing method | |
| Vinogradov et al. | Robust multiphysics optimization for fan blade aerodynamic efficiency, structural properties and flutter sensitivity | |
| CN115481511A (en) | Centrifugal impeller multi-working-condition local configuration pneumatic optimization method and device based on FFD | |
| Vasilopoulos et al. | CAD-based aerodynamic optimization of a compressor stator using conventional and adjoint-driven approaches | |
| Gao et al. | A robust blade design method based on non-intrusive polynomial chaos considering profile error | |
| Kirz | Surrogate based shape optimization of a low boom fuselage wing configuration | |
| CN114861315A (en) | Two-dimensional impeller profile optimization method based on machine learning | |
| Luers et al. | Adjoint-based volumetric shape optimization of turbine blades | |
| CN114861315B (en) | Turbine two-dimensional blade profile optimization method based on machine learning | |
| Meng et al. | A nurbs-enhanced finite volume method for steady euler equations with goal-oriented h-adaptivity | |
| CN114065572B (en) | Blade model construction method related to thickness deviation and blade thickness distribution | |
| CN110704944A (en) | Variable camber airfoil profile-oriented parametric modeling method | |
| CN114880792A (en) | Deformation prediction-based omnibearing multi-angle optimization method | |
| CN113486512A (en) | Flutter analysis method for functional gradient variable-thickness blade model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |