CN116399541A - Blade grid wind tunnel experiment working condition parameter correction method based on deep neural network - Google Patents
Blade grid wind tunnel experiment working condition parameter correction method based on deep neural network Download PDFInfo
- Publication number
- CN116399541A CN116399541A CN202310354263.1A CN202310354263A CN116399541A CN 116399541 A CN116399541 A CN 116399541A CN 202310354263 A CN202310354263 A CN 202310354263A CN 116399541 A CN116399541 A CN 116399541A
- Authority
- CN
- China
- Prior art keywords
- parameters
- flow field
- neural network
- experimental
- flow
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 65
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 43
- 238000012937 correction Methods 0.000 title claims abstract description 22
- 238000002474 experimental method Methods 0.000 title abstract description 8
- 238000004364 calculation method Methods 0.000 claims abstract description 23
- 238000005259 measurement Methods 0.000 claims abstract description 22
- 238000004088 simulation Methods 0.000 claims abstract description 19
- 238000005457 optimization Methods 0.000 claims abstract description 14
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 8
- 238000012986 modification Methods 0.000 claims description 7
- 230000004048 modification Effects 0.000 claims description 7
- 230000004069 differentiation Effects 0.000 claims description 4
- 239000004973 liquid crystal related substance Substances 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 230000003068 static effect Effects 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 230000009191 jumping Effects 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 2
- 238000000926 separation method Methods 0.000 abstract description 9
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 238000001914 filtration Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M9/00—Aerodynamic testing; Arrangements in or on wind tunnels
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M9/00—Aerodynamic testing; Arrangements in or on wind tunnels
- G01M9/02—Wind tunnels
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M9/00—Aerodynamic testing; Arrangements in or on wind tunnels
- G01M9/02—Wind tunnels
- G01M9/04—Details
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M9/00—Aerodynamic testing; Arrangements in or on wind tunnels
- G01M9/06—Measuring arrangements specially adapted for aerodynamic testing
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M9/00—Aerodynamic testing; Arrangements in or on wind tunnels
- G01M9/06—Measuring arrangements specially adapted for aerodynamic testing
- G01M9/065—Measuring arrangements specially adapted for aerodynamic testing dealing with flow
- G01M9/067—Measuring arrangements specially adapted for aerodynamic testing dealing with flow visualisation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M9/00—Aerodynamic testing; Arrangements in or on wind tunnels
- G01M9/08—Aerodynamic models
-
- 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
- Physics & Mathematics (AREA)
- Fluid Mechanics (AREA)
- General Physics & Mathematics (AREA)
- Aerodynamic Tests, Hydrodynamic Tests, Wind Tunnels, And Water Tanks (AREA)
Abstract
The invention relates to a cascade wind tunnel experimental condition parameter correction method based on a deep neural network, and belongs to the field of aerospace experimental measurement and artificial intelligence. The floating working condition parameters are used for constructing a plurality of groups of incoming flow parameters and carrying out the numerical calculation of the cascade flow field; taking normalized floating working condition parameters as input quantity, taking normalized corresponding physical quantity of measuring point positions in the region 50% in front of the chord length of the blade of the numerical flow field as output, and constructing a deep neural network; taking the mean square error of the prediction result and the experimental measurement result of the deep neural network as a loss function, and using an automatic differential algorithm to counter-propagate the gradient of the loss function to obtain the gradient of the mean square error to the working condition parameters; and correcting the working condition parameters by using a gradient optimization algorithm. The method can effectively correct the incoming flow boundary condition of the cascade wind tunnel experiment, can effectively overcome the prediction deviation of the separation area generated by the turbulence model, does not need to specially select the turbulence model, and is particularly suitable for reconstructing and inverting the experimental flow field by using a flow field numerical simulation method.
Description
Technical Field
The invention belongs to the field of aerospace experimental measurement and artificial intelligence, and relates to a blade grid wind tunnel experimental condition parameter correction method.
Background
The cascade wind tunnel experiment is a common method for measuring the blade profile performance of the impeller machinery, and the working condition parameters of the cascade wind tunnel experiment mainly comprise incoming flow attack angle and incoming flow Mach number: the incoming flow attack angle is generally obtained by rotating the cascade mounting turntable to a corresponding angle; the Mach number of the incoming flow is displayed on a Mach number sensor at the front part of the experimental part by adjusting the rotation speed of a valve or a fan of a wind tunnel air source. However, in the actual wind tunnel experimental process, the airflow in the wind tunnel is influenced by the wall surface of the wind tunnel, the loss and the deflection of the airflow exist after the airflow flows through the Mach number sensor, and the Mach number and the attack angle of the airflow truly blown to the blade grid area are not given working condition parameters. At a large attack angle and a high Mach number, the working condition deviation is more obvious, and the effect of the blade grid wind tunnel experiment is seriously influenced.
The existing working condition parameter correction method adopts an integrated Kalman filtering method, but the integrated Kalman filtering method is a quasi-linear method, nonlinear information of a flowing process in a wind tunnel cannot be captured, correction accuracy and application range are limited, and the neural network has strong nonlinear fitting capability, can effectively capture the nonlinear information, and has high modeling accuracy.
Therefore, the current cascade wind tunnel experimental field lacks an accurate and reliable working condition parameter correction method.
Disclosure of Invention
The technical problems to be solved by the invention are as follows:
in order to overcome the defect that nonlinear information of a flowing process in a wind tunnel cannot be captured by adopting a set Kalman filtering method in the prior art, the invention provides a blade grid wind tunnel experimental condition parameter correction method based on flow field numerical simulation and a deep neural network.
In order to solve the technical problems, the invention adopts the following technical scheme:
a blade grid wind tunnel experimental condition parameter correction method based on flow field numerical simulation and a deep neural network is characterized by comprising the following steps:
step 1: floating experimental working condition parameters, constructing a plurality of groups of incoming flow parameter samples, and carrying out flow field numerical simulation on each group of parameters;
step 2: constructing a neural network describing the incoming flow parameters and the flow field parameters, wherein the input parameters of the neural network are normalized incoming flow parameter samples in the step 1, and the output parameters are normalized corresponding physical quantities of measurement positions of 50% of the front area of the chord length of the blade in the flow field calculated by each sample;
step 3: taking the experimental measured working condition parameters as initial values; correcting incoming flow parameters by using a gradient optimization method, and updating flow field input parameters:
step 4: the flow field input parameters obtained by current calculation are used as working condition parameters after modification of the experimental working condition of the modified blade cascade, flow field numerical calculation is carried out by using the Mach number of the modified incoming flow and the incoming flow attack angle, and if the mean square error of the flow parameters of the measuring points in the numerical flow field and experimental measurement data of the first 50% chord length area is smaller than a specified threshold, the modification parameters can be used as final modification results; and if the mean square error is greater than the specified threshold, taking the flow field numerical value calculation data as a training set added into the deep neural network, and repeating the steps 1 to 4.
The invention further adopts the technical scheme that: the step 1 is specifically as follows: and setting a floating range according to the regulation precision of the incoming flow parameters, randomly sampling by using a Latin hypercube method, constructing samples of the incoming flow parameters, and carrying out cascade flow field numerical calculation on each sample in a turbulence model.
The invention further adopts the technical scheme that: the corresponding physical quantities described in step 2 include incoming flow Mach number, pressure coefficient, static pressure.
The invention further adopts the technical scheme that: the step 2 is specifically as follows:
step 2-1: the input parameters of the neural network are normalizedThe incoming parameters of each sample after the conversion are equal to the incoming Mach number Ma in And angle of attack α, the normalization formula is as follows:
wherein alpha is max 、α min 、Ma in，max 、Ma in，min Respectively representing the maximum value and the minimum value of incoming flow attack angles and incoming flow Mach numbers in training data;
step 2-2: the output parameters of the neural network are physical quantities of blade measuring point positions corresponding to the first 50% chord length in the flow field calculation result of each sample after normalization, and the physical quantity of the jth measuring point
The normalization formula is as follows:
wherein j=1, 2, …, M, M represents the number of blade measuring points corresponding to the first 50% chord length in the experimental process,
respectively the minimum value and the maximum value of the physical quantity of the jth measuring point in the sample flow field;
step 2-3: the method comprises the steps of constructing a deep neural network DNN, wherein the deep neural network DNN comprises an input layer, a plurality of hidden layers and an output layer, and the mathematical expression is as follows:
the invention further adopts the technical scheme that: the step 3 is specifically as follows:
step 3-1: selecting experimental working condition parameters as initial values Ma in,0 ,α 0 ；
Step 3-2: for the kth step, k=0, 1,2, …, k max ,k max For maximum iteration times, performing neural network prediction, and inputting parameters Ma in,k And alpha k Normalizing to obtain DNN output result, and inversely transforming according to formula (3) to obtain physical quantity of corresponding position of measuring point
And calculating experimental measurement data of corresponding measuring points>
Is the mean square error of the loss function J (Ma in,k ,α k ) Abbreviated as J k
Wherein, the liquid crystal display device comprises a liquid crystal display device,
representing the measured physical quantity of the jth measuring point in the experimental measurement result, if J k+1 <Epsilon or k equal to k max Jumping to step 4, wherein epsilon is a specified threshold;
step 3-3: using an automatic differentiation method, for the loss function J k Performing gradient back propagation to obtain gradient of loss function to incoming flow parameters
Step 3-4: correcting incoming flow parameters by using a gradient optimization method, and updating flow field input parameters:
opt is a gradient optimization algorithm, and the updated flow field input parameters are brought into the step 3-2 for iteration.
The invention further adopts the technical scheme that: the gradient optimization method comprises a gradient descent method, adam and AdaGrad.
A computer system, comprising: one or more processors, a computer-readable storage medium storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the methods described above.
A computer readable storage medium, characterized by storing computer executable instructions that when executed are configured to implement the method described above.
The invention has the beneficial effects that:
the blade grid wind tunnel experimental condition parameter correction method based on the flow field numerical simulation and the deep neural network provided by the invention can effectively correct the incoming flow boundary condition of the blade grid wind tunnel experiment, can effectively overcome the prediction deviation of a separation area generated by a turbulence model in the flow field numerical simulation in the correction process, does not need to specially select the turbulence model, and is particularly suitable for reconstructing and inverting the experimental flow field by using the flow field numerical simulation method. The method is simple, and the accuracy of the correction result is high.
The method comprises the following steps:
1. because flow separation generally occurs in the area of 50% behind the blade, the turbulence model prediction flow separation used in the current flow field numerical simulation is not very accurate, so that 50% of flow field prediction data behind the blade is polluted by turbulence model prediction deviation, and correction errors can be caused. Thus, the present invention uses data for the first 50% chord length region rather than all data, while the first 50% region is an attached flow without flow separation, and turbulence models can generally be predicted very accurately.
2. Compared with the existing ensemble Kalman filtering method, the method provided by the invention has the advantages that the nonlinear information of the flowing process in the wind tunnel can be better captured, the modeling is more accurate, and the correction precision is higher.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, like reference numerals being used to refer to like parts throughout the several views.
FIG. 1 is a flow chart of a flow field numerical simulation and deep neural network-based cascade wind tunnel experimental condition parameter correction method;
fig. 2 is a graph comparing flow field calculation results before and after the correction of the incoming flow parameters with experimental measurement results.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
The invention provides a cascade wind tunnel experimental condition parameter correction method based on flow field numerical simulation and a deep neural network, which is used for floating incoming flow Mach number and attack angle of experimental conditions to be corrected, constructing a plurality of groups of incoming flow parameters and calculating cascade flow field numerical values; taking normalized incoming flow Mach number and attack angle after floating as input quantity, taking normalized corresponding physical quantity of measuring point position of 50% area before the chord length of the blade of the numerical flow field as output, and constructing a deep neural network; taking the mean square error of the prediction result and the experimental measurement result of the deep neural network as a loss function, and using an automatic differential algorithm to counter-propagate the gradient of the loss function to obtain the gradient of the mean square error on the Mach number of the incoming flow and the attack angle of the incoming flow; correcting the incoming flow Mach number and the incoming flow attack angle by using a gradient optimization algorithm; and carrying out flow field numerical calculation by using the corrected incoming flow Mach number and incoming flow attack angle, if the mean square error of the flow parameters of the measuring points in the numerical flow field and the experimental measurement result is smaller than a specified threshold value, using the corrected parameters as a final correction result, and if the mean square error is larger than the specified threshold value, using the flow field numerical calculation data as a training set added into the first-step deep neural network, and repeating the flow.
As shown in fig. 1, the method comprises the following steps:
step 1: the floating working condition parameters, the specific floating amplitude depends on the regulation precision of the incoming flow parameters, and a typical floating range can be set as follows: mach number of incoming stream Ma in Floating + -0.05, floating + -1 DEG of incoming flow attack angle alpha, randomly sampling by using Latin hypercube method, constructing a plurality of samples of incoming flow parameters, selecting a proper turbulence model method, and calculating the value of the cascade flow field for each sample;
step 2: constructing a neural network describing the incoming flow parameters and the flow field parameters, wherein the input parameters of the neural network are normalized incoming flow parameter samples in the step 1, the output parameters are normalized corresponding physical quantities of measuring positions of 50% of the front area of the chord length of the blade in the flow field calculated for each sample, and the measured physical quantities include but are not limited to: surface isentropic Mach number, pressure coefficient, static pressure, etc.; the data of the first 50% chord length area is adopted instead of all the data, because flow separation generally occurs in the area of the first 50% of the rear part of the blade, the current turbulence model prediction flow separation is not quite accurate, so that the data of the second 50% of the blade is polluted by the turbulence model prediction deviation, and a correction error is caused, and the area of the first 50% of the blade is attached flow without flow separation, and the turbulence model can generally predict very accurately.
Step 3: taking the experimental measured working condition parameters as initial values; correcting incoming flow parameters by using a gradient optimization method, and updating flow field input parameters:
wherein:J k as a loss function
In the method, in the process of the invention,
for the physical quantity measured at the jth measuring point during the experiment,/>
For the result of the j-th measuring point predicted by the deep neural network, j=1, 2, …, M and M represent the number of measuring points in the front 50% chord length area of the blade in the experimental process, and Ma in,k ，α k For the incoming flow parameter of the kth iteration, k=0, 1,2, …, k max ,k max For maximum iteration number, the automatic differentiation method is used for the loss function J k Gradient counter-propagation is carried out to obtain the gradient of the loss function to the incoming flow parameters>
Step 4: the flow field input parameters obtained by current calculation are used as working condition parameters after modification of the experimental working condition of the modified blade cascade, flow field numerical calculation is carried out again, experimental measuring point corresponding data of a calculation result is compared with experimental measurement data (the first 50% chord length area), if deviation is obviously reduced compared with the prior art, the deviation can be more than 50% after correction, the working condition parameters are output results, and the algorithm is ended; if the deviation is not obviously reduced, adding the calculated data into the training sample in the step 1, and carrying out the steps 1 to 4 again.
Example 1:
taking a MAN GHH 1-S1 compressor cascade as an example, the design Mach number is 0.62, the design attack angle is 0 degrees, the distribution of the isentropic Mach numbers of the surfaces of 10 measuring points of the pressure surface and the suction surface of the blade is experimentally measured, the incoming flow Mach number and the incoming flow attack angle of the experimental result of the wind tunnel under the design working condition are corrected, the numerical simulation and experimental measurement result of the isentropic Mach number of the surface of the blade before the parameters are corrected are shown in figure 2, and the specific implementation steps of the embodiment are as follows:
step 1: the floating working condition parameters, the specific floating amplitude depends on the regulation precision of the incoming flow parameters, and a typical floating range can be set as follows: the Mach number of the floating incoming flow is 0.62+/-0.05, the incoming flow attack angle floats 0+/-1 DEG, random sampling can be carried out by using atin hypercube method and the like, samples of incoming flow parameters are constructed, the number of the samples is taken as 32, and the numerical calculation of the cascade flow field is carried out on each sample.
Step 2: the method comprises the following specific steps of:
step 2-1: the input parameters of the neural network are normalized incoming flow Mach number and incoming flow attack angle of each sample, and the incoming flow Mach number Ma in And angle of attack α, the normalization formula is as follows:
wherein alpha is max 、α min 、Ma in，max 、Ma in，min Respectively representing the maximum value and the minimum value of incoming flow attack angles and incoming flow Mach numbers in training data;
step 2-2: the output parameters of the neural network are the surface isentropic Mach numbers of the blade measuring point positions corresponding to the first 50% chord length in the flow field calculation result of each sample after normalization, and the surface isentropic Mach numbers Ma of j measuring points is,j The normalization formula is as follows:
where j=1, 2,..m, M represents the number of blade stations corresponding to the first 50% chord length region in the experimental measurement, in this example m=12, ma is,j,min 、Ma is,j,max The surface isentropic Mach numbers of the j-th measuring point in the sample flow field are the minimum value and the maximum value of the surface isentropic Mach numbers of the j-th measuring point, in the example, the surface isentropic Mach numbers of the pressure surface and the suction surface of the blade are experimentally measured, and in practical application, the physical quantity measured on the surface of the blade can be: surface isentropic Mach number, pressure coefficient, static pressure, etc.;
step 2-3: for the fully connected network of the embodiment, one input layer and one output layer, three hidden layers basically meet the precision requirement, and the mathematical expression is as follows:
step 3: taking the experimental measured working condition parameters as initial values; the gradient descent optimization method is used for correcting incoming flow parameters and updating flow field input parameters, and the specific steps are as follows:
step 3-1: for this example, the operating mode parameter initial value is Ma in,0 ＝0.62，α 0 ＝0°；
Step 3-2: for the k-th step, k=0, 1,2,.. max ,k max For maximum iteration times, performing neural network prediction, and inputting parameters Ma in,k And alpha k Normalizing according to formulas (1) and (2) respectively to obtain DNN output results, and performing inverse transformation according to formula (3) to obtain isentropic Mach number Ma of the surface of the corresponding position of the measuring point is,j And calculate the experimental measurement data Ma of the corresponding measuring point is,exp,j Is the mean square error of the loss function J (Ma in,k ,α k ) Abbreviated as J k
Wherein Ma is,exp,j Indicating the isentropic Mach number of the surface of the jth measuring point in the experimental measurement result, if J k+1 <Epsilon or k equal to k max Jumping to step 4, wherein epsilon is a specified threshold;
step 3-3: using an automatic differentiation method, for the loss function J k Performing gradient back propagation to obtain gradient of loss function to incoming flow parameters
Step 3-4: correcting incoming flow parameters by using a gradient optimization method, such as a gradient descent method, adam, adaGrad and the like, and updating flow field input parameters:
opt is a gradient optimization algorithm, the updated flow field input parameters are brought into the step 3-2 for iteration, and the calculation time is very small because the prediction uses a deep neural network;
step 4: the flow field input parameters obtained by current calculation are used as working condition parameters after the modification of the experimental working condition of the modified blade cascade, the flow field numerical calculation is carried out again, experimental measurement point corresponding data of a calculation result is compared with experimental measurement data of a first 50% chord length area, if deviation is obviously reduced compared with the previous one, the reduction amplitude is determined according to specific conditions, and the working condition parameters are output results and the algorithm is ended after the correction can be generally carried out by about 50%; if the deviation is not obviously reduced, adding the calculated data into the training sample in the step 1, and carrying out the steps 1 to 4 again. For the present example and most cases, the satisfactory results are obtained by performing steps 1 to 4 once. In this example, the corrected operating parameters are: the incoming flow Mach number 0.6369, the incoming flow attack angle is-0.3259 °.
The comparison graph of the isentropic Mach number curve of the corrected predicted result and the experimental measurement result is shown in the attached figure 2, and it can be seen that after the incoming flow parameter is corrected, the flow field numerical simulation result well accords with the experimental result except for the trailing edge region, and the deviation exists in the trailing edge region because the turbulence model has limited prediction capability on the separation region, but the turbulence model has higher accuracy on the attached flow prediction.
While the invention has been described with reference to certain preferred embodiments, it will be understood by those skilled in the art that various changes and substitutions of equivalents may be made without departing from the spirit and scope of the invention.
Claims (8)
1. A blade grid wind tunnel experimental condition parameter correction method based on flow field numerical simulation and a deep neural network is characterized by comprising the following steps:
step 1: floating experimental working condition parameters, constructing a plurality of groups of incoming flow parameter samples, and carrying out flow field numerical simulation on each group of parameters;
step 2: constructing a neural network describing the incoming flow parameters and the flow field parameters, wherein the input parameters of the neural network are normalized incoming flow parameter samples in the step 1, and the output parameters are normalized corresponding physical quantities of measurement positions of 50% of the front area of the chord length of the blade in the flow field calculated by each sample;
step 3: taking the experimental measured working condition parameters as initial values; correcting incoming flow parameters by using a gradient optimization method, and updating flow field input parameters:
step 4: and (3) taking the current calculation to obtain flow field input parameters as working condition parameters after modification of the experimental working condition of the modified blade cascade, carrying out flow field numerical calculation by using the modified incoming flow Mach number and incoming flow attack angle, if the mean square error of the flow parameters of the measuring points in the numerical flow field and experimental measurement data of the former 50% chord length area is smaller than a specified threshold, taking the modified parameters as a final modification result, and if the mean square error is larger than the specified threshold, taking the flow field numerical calculation data as a training set added into the deep neural network, and repeating the steps (1) to (4).
2. The method for correcting the experimental condition parameters of the cascade wind tunnel based on the flow field numerical simulation and the deep neural network according to claim 1, wherein the step 1 is specifically as follows: and setting a floating range according to the regulation precision of the incoming flow parameters, randomly sampling by using a Latin hypercube method, constructing samples of the incoming flow parameters, and carrying out cascade flow field numerical calculation on each sample in a turbulence model.
3. The method for correcting the experimental condition parameters of the cascade wind tunnel based on the flow field numerical simulation and the deep neural network according to claim 1, wherein the corresponding physical quantities in the step 2 comprise incoming flow Mach number, pressure coefficient and static pressure.
4. The method for correcting the experimental condition parameters of the cascade wind tunnel based on the flow field numerical simulation and the deep neural network according to claim 2, wherein the step 2 is specifically as follows:
step 2-1: the input parameters of the neural network are normalized incoming flow parameters of each sample, and the incoming flow Mach number Ma is calculated in And angle of attack α, the normalization formula is as follows:
wherein alpha is max 、α min 、Ma in，max 、Ma in，min Respectively representing the maximum value and the minimum value of incoming flow attack angles and incoming flow Mach numbers in training data;
step 2-2: the output parameters of the neural network are physical quantities of blade measuring point positions corresponding to the first 50% chord length in the flow field calculation result of each sample after normalization, and the physical quantity of the jth measuring point
The normalization formula is as follows:
wherein j=1, 2, …, M, M represents the number of blade measuring points corresponding to the first 50% chord length in the experimental process,
respectively the minimum value and the maximum value of the physical quantity of the jth measuring point in the sample flow field;
step 2-3: the method comprises the steps of constructing a deep neural network DNN, wherein the deep neural network DNN comprises an input layer, a plurality of hidden layers and an output layer, and the mathematical expression is as follows:
5. the method for correcting the experimental condition parameters of the cascade wind tunnel based on the flow field numerical simulation and the deep neural network according to claim 4, wherein the step 3 is specifically as follows:
step 3-1: selecting experimental working condition parameters as initial values Ma in,0 ,α 0 ；
Step 3-2: for the kth step, k=0, 1,2, …, k max ,k max For maximum iteration times, performing neural network prediction, and inputting parameters Ma in,k And alpha k Normalizing to obtain DNN output result, and inversely transforming according to formula (3) to obtain physical quantity of corresponding position of measuring point
And calculating experimental measurement data of corresponding measuring points>
Is the mean square error of the loss function J (Ma in,k ,α k ) Abbreviated as J k
Wherein, the liquid crystal display device comprises a liquid crystal display device,
representing the measured physical quantity of the jth measuring point in the experimental measurement result, if J k+1 <Epsilon or k equal to k max Jumping to step 4, wherein epsilon is a specified threshold;
step 3-3: using an automatic differentiation method, for the loss function J k Performing gradient back propagation to obtain gradient of loss function to incoming flow parameters
Step 3-4: correcting incoming flow parameters by using a gradient optimization method, and updating flow field input parameters:
opt is a gradient optimization algorithm, and the updated flow field input parameters are brought into the step 3-2 for iteration.
6. The method for correcting the experimental condition parameters of the cascade wind tunnel based on the flow field numerical simulation and the deep neural network according to claim 5, wherein the gradient optimization method comprises a gradient descent method, adam and adaGrad.
7. A computer system, comprising: one or more processors, a computer-readable storage medium storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of claim 1.
8. A computer readable storage medium, characterized by storing computer executable instructions that, when executed, are adapted to implement the method of claim 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310354263.1A CN116399541A (en) | 2023-04-04 | 2023-04-04 | Blade grid wind tunnel experiment working condition parameter correction method based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310354263.1A CN116399541A (en) | 2023-04-04 | 2023-04-04 | Blade grid wind tunnel experiment working condition parameter correction method based on deep neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116399541A true CN116399541A (en) | 2023-07-07 |
Family
ID=87011780
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310354263.1A Pending CN116399541A (en) | 2023-04-04 | 2023-04-04 | Blade grid wind tunnel experiment working condition parameter correction method based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116399541A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116907880A (en) * | 2023-09-13 | 2023-10-20 | 中汽研新能源汽车检验中心(天津)有限公司 | Air supply equipment for testing vehicle and air supply control method |
| CN117313579A (en) * | 2023-10-07 | 2023-12-29 | 中国航空发动机研究院 | Engine compression part flow field prediction method, device, equipment and storage medium |
-
2023
- 2023-04-04 CN CN202310354263.1A patent/CN116399541A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116907880A (en) * | 2023-09-13 | 2023-10-20 | 中汽研新能源汽车检验中心(天津)有限公司 | Air supply equipment for testing vehicle and air supply control method |
| CN116907880B (en) * | 2023-09-13 | 2023-11-17 | 中汽研新能源汽车检验中心(天津)有限公司 | Air supply equipment for testing vehicle and air supply control method |
| CN117313579A (en) * | 2023-10-07 | 2023-12-29 | 中国航空发动机研究院 | Engine compression part flow field prediction method, device, equipment and storage medium |
| CN117313579B (en) * | 2023-10-07 | 2024-04-05 | 中国航空发动机研究院 | Engine compression part flow field prediction method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116399541A (en) | Blade grid wind tunnel experiment working condition parameter correction method based on deep neural network | |
| CN104361414B (en) | Power transmission line icing prediction method based on correlation vector machine | |
| CN110672328B (en) | Turbofan engine health parameter estimation method based on random configuration network | |
| CN105718634B (en) | A kind of aerofoil profile Robust Optimal Design based on non-probability interval analysis model | |
| CN111027148B (en) | Automatic calibration and industrial axial flow compressor performance calculation method for loss lag angle model | |
| CN112632719B (en) | Multi-stage axial flow compressor characteristic correction method based on one-dimensional average flow line method | |
| CN110119588A (en) | On-line optimization design method based on Extended Kalman filter state estimation | |
| CN110765706B (en) | Aerofoil unsteady stall aerodynamic coefficient modeling method based on OHNGBM (1, 1) | |
| CN111780949B (en) | CFD analysis-based total pressure correction method for high-speed air inlet channel precursor wind tunnel experiment | |
| CN113947035A (en) | Data heaven-earth correlation method for transition of hypersonic velocity boundary layer | |
| CN116796141A (en) | GBDT regression model-based office building energy consumption prediction method | |
| CN114742307A (en) | Wave element prediction method and system | |
| CN114154275B (en) | Low-pressure turbine blade profile pneumatic design method based on optimal load distribution model optimization | |
| Kun et al. | Nested sparse-grid stochastic collocation method for uncertainty quantification of blade stagger angle | |
| Wang et al. | Effects of leading edge defect on the aerodynamic and flow characteristics of an S809 airfoil | |
| Huppertz et al. | Knowledge based 2D blade design using multi-objective aerodynamic optimization and a neural network | |
| CN113095547B (en) | Short-term wind power prediction method based on GRA-LSTM-ICE model | |
| CN114154234A (en) | Modeling method, system and storage medium for aircraft engine | |
| CN112215281A (en) | Fan blade icing fault detection method | |
| CN112685954B (en) | Method and device for predicting wind speed and fan rotating speed of automobile environment wind tunnel | |
| CN116085245A (en) | Online compressor performance prediction method and system based on OS-ELM | |
| Wilkinson et al. | A comparison of actuator disc models for axial flow fans in large air-cooled heat exchangers | |
| CN112541308A (en) | Novel aerospace vehicle FADS system pressure measuring hole layout optimization method | |
| CN115081100A (en) | Training method of nacelle design model, and manufacturing method and system of nacelle | |
| CN116399542A (en) | Blade grid wind tunnel experiment working condition parameter correction method based on average set Kalman filtering |
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 |