Neural network correction method for radiation error of turbine blade modeling theory
Through the neural network correction method, the condition generation adversarial network predicts the cooling effect of the turbine blades under low temperature conditions under low temperature conditions, solving the problem of inefficient cooling efficiency evaluation caused by radiation error in the prior art, and realizing the cooling efficiency under high temperature conditions Accurate cold efficiency evaluation and improve calculation efficiency.
Patent Information
- Application Number
- CN202411359968.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-27
- Publication Date
- 2025-01-10
AI Technical Summary
The existing modulation theory fails to effectively consider radiation factors when evaluating the cooling efficiency of turbine blades, resulting in low computational efficiency and it is difficult to accurately evaluate the cooling efficiency under high temperature conditions under low temperature conditions.
The neural network correction method is adopted to build a condition-generating adversarial network, and the cold effect data under low temperature conditions and the modular condition parameters are used to predict the cold effect under high temperature conditions, so as to improve the cold effect evaluation accuracy and calculation efficiency.
It significantly improves the accuracy and calculation efficiency of engine turbine blade cooling efficiency evaluation, can accurately predict the cooling efficiency under high temperature conditions under low temperature conditions, and enhances the applicability of the molding theory.
Smart Images
Figure CN119293992A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of aero-engine flow and heat transfer analysis, and in particular relates to a neural network correction method for theoretical radiation errors of turbine blade modeling. Background Art
[0002] The working environment temperature of the turbine blades of aircraft engines far exceeds the temperature resistance value of its own materials. Therefore, various efficient cooling measures are used for the blades in practical applications. The cooling effect of the cooling measures can be measured by the cooling efficiency, which can be obtained through experiments. However, it is very difficult to restore the real high-temperature environment of turbine blades in the laboratory. The existing modeling theory can control the experimental temperature under medium and low temperature conditions, and the cooling effect can be evaluated by measuring the cooling effect under medium and low temperature conditions. However, the existing modeling theory does not take radiation into account and has low calculation efficiency. Summary of the invention
[0003] The purpose of the present invention is to provide a neural network correction method for the theoretical radiation error of the modeling of aircraft engine turbine blades. By adding the influence of high temperature radiation to the cooling effect under low temperature conditions, the cooling effect under high temperature conditions can be quickly obtained, which significantly improves the cooling effect evaluation accuracy of the engine turbine blades and greatly improves the calculation efficiency.
[0004] To achieve the above object, the present invention provides a neural network correction method for the theoretical radiation error of turbine blade modeling, comprising:
[0005] Constructing a turbine blade cooling data set, building a conditional generative adversarial network according to the sizes of cooling data and operating condition data in the turbine blade cooling data set, and training the conditional generative adversarial network to obtain a target conditional generative adversarial network;
[0006] Based on the target condition generative adversarial network, the cooling effect of high temperature conditions is predicted according to the cooling effect of low temperature conditions and the operating parameters of the simulation process, and the accuracy and generalization of the target condition generative adversarial network are verified.
[0007] Preferably, the process of constructing a turbine blade cooling effect data set includes:
[0008] Based on the turbine blade modeling theory, a dataset of turbine blade cooling effect under radiation is constructed;
[0009] The data samples in the turbine blade cooling effect data set include cooling effect under low temperature conditions, cooling effect under high temperature conditions and simulation condition parameters.
[0010] Preferably, based on the turbine blade modeling theory, the process of constructing a turbine blade cooling effect data set under radiation includes:
[0011] The cooling effect training data is obtained by calculating the CFD numerical value through the calculation model of the turbine blade film cooling unit, and a modeling process is constructed based on the cooling effect training data to obtain the turbine blade cooling effect data set.
[0012] Preferably, the process of constructing a modeling process based on the cold effect training data comprises:
[0013] Two working conditions with different gas temperatures are modeled from low temperature to high temperature, and the medium gas temperature is used as both the low temperature in one set of modeling processes and the high temperature in another set of modeling conditions. The modeling process is constructed by the mutual combination of gas temperatures.
[0014] Preferably, the calculating CFD values involves CFD calculation conditions;
[0015] The variables selected for the CFD calculation conditions include: wall emissivity, absorption coefficient, and gas temperature.
[0016] Preferably, the CFD numerical CFD calculation result is the temperature field distribution of the cooling unit. When constructing the turbine blade cooling effect data set, the temperature field distribution is converted into cooling effect data. The formula expression is:
[0017]
[0018] Among them, T g is the gas temperature, T aw is the wall temperature under adiabatic conditions, T c The air conditioning temperature.
[0019] Preferably, an L1 regularization term is added to the objective function used to construct the conditional generative adversarial network, which is expressed as:
[0020]
[0021] Among them, G is the generator, D is the discriminator, x is the operating condition data, y is the cooling effect under high temperature conditions, and λ is the regularization factor.
[0022] Preferably, the conditional generative adversarial network includes a generator and a discriminator;
[0023] The generator is a U-Net with a residual network structure added;
[0024] The discriminator is a binary classifier, which is trained in conjunction with the generator. The input of the discriminator is cold effect data, and the output is the attribute of the cold effect.
[0025] Preferably, the input of the target condition generation adversarial network is the cooling efficiency data under low temperature conditions and the simulation condition data, and the output is the cooling efficiency data under high temperature conditions.
[0026] Preferably, the process of verifying the accuracy and generalization of the target conditional generative adversarial network includes:
[0027] The turbine blade cooling effect data set is divided into a verification data set and a training data set. The cooling effect under low-temperature conditions and the working conditions of the simulation process in the verification data set are input into the generator in turn, and the generated cooling effect under high-temperature conditions is compared with the actual results in the verification data set to verify the accuracy and generalization of the network.
[0028] Compared with the prior art, the present invention has the following advantages and technical effects:
[0029] The present invention discloses a neural network correction method for the theoretical radiation error of the modeling of aero-engine turbine blades, which is used to predict the air film cooling effect under high-temperature conditions. The distribution characteristics of the cooling effect under low-temperature conditions are extracted by using a neural network, and the high-temperature radiation effect is added to the cooling effect under low-temperature conditions according to the working condition data, so as to obtain accurate high-temperature cooling effect. The present invention can significantly improve the cooling effect evaluation accuracy of engine turbine blades and greatly improve the calculation efficiency, and has broad application prospects.
[0030] The present invention proposes a neural network correction method for the radiation error of the modeling theory of aircraft engine turbine blades. It only needs to input the cooling effect under low-temperature conditions and the operating parameters of the modeling process into the network. The network outputs the cooling effect under the target high-temperature condition through calculation, thereby increasing the accuracy of the modeling theory in predicting the cooling effect, improving the temperature assessment accuracy of the engine turbine blades, and greatly improving the calculation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0032] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION
[0033] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0034] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0035] like Figure 1As shown, this embodiment provides a neural network correction method for the theoretical radiation error of turbine blade modeling, including:
[0036] Construct a turbine blade cooling data set, build a conditional generative adversarial network based on the size of the cooling data and operating condition data in the turbine blade cooling data set, and train the conditional generative adversarial network to obtain a target conditional generative adversarial network;
[0037] Based on the target condition generative adversarial network, the cooling effect of high temperature conditions is predicted according to the cooling effect of low temperature conditions and the operating parameters of the simulation process, and the accuracy and generalization of the target condition generative adversarial network are verified.
[0038] Furthermore, the process of constructing the turbine blade cooling effect dataset includes:
[0039] Based on the turbine blade modeling theory, a dataset of turbine blade cooling effect under radiation is constructed;
[0040] The data samples in the turbine blade cooling effect data set include cooling effect under low temperature conditions, cooling effect under high temperature conditions and simulation condition parameters.
[0041] Furthermore, based on the turbine blade modeling theory, the process of constructing a turbine blade cooling effect dataset under radiation includes:
[0042] The cooling effect training data is obtained by calculating the CFD numerical value through the calculation model of the turbine blade film cooling unit. The modeling process is constructed based on the cooling effect training data to obtain the turbine blade cooling effect data set.
[0043] Furthermore, the process of constructing a modeling process based on the cold effect training data includes:
[0044] Two working conditions with different gas temperatures are modeled from low temperature to high temperature, and the medium gas temperature is used as both the low temperature in one set of modeling processes and the high temperature in another set of modeling conditions. The modeling process is constructed by the mutual combination of gas temperatures.
[0045] Furthermore, calculating CFD values involves CFD calculation conditions;
[0046] The variables selected for CFD calculation conditions include: wall emissivity, absorption coefficient, and gas temperature.
[0047] Furthermore, the CFD numerical calculation result is the temperature field distribution of the cooling unit. When constructing the turbine blade cooling effect data set, the temperature field distribution is converted into cooling effect data. The formula expression is:
[0048]
[0049] Among them, T g is the gas temperature, Taw is the wall temperature under adiabatic conditions, T c The air conditioning temperature.
[0050] Furthermore, an L1 regularization term is added to the objective function used to construct the conditional generative adversarial network, which is expressed as:
[0051]
[0052] Among them, G is the generator, D is the discriminator, x is the operating condition data, y is the cooling effect under high temperature conditions, and λ is the regularization factor.
[0053] Furthermore, the conditional generative adversarial network includes a generator and a discriminator;
[0054] The generator is U-Net with a residual network structure added;
[0055] The discriminator is a binary classifier, which is trained with the generator. The input of the discriminator is the cold effect data, and the output is the attribute of the cold effect.
[0056] Furthermore, the input of the target condition generative adversarial network is the cooling efficiency data under low temperature conditions and the simulation condition data, and the output is the cooling efficiency data under high temperature conditions.
[0057] Furthermore, the process of verifying the accuracy and generalization of the target-conditional generative adversarial network includes:
[0058] The turbine blade cooling effect dataset is divided into a validation dataset and a training dataset. The cooling effect under low temperature conditions and the working conditions of the simulation process in the validation dataset are input into the generator in turn. The generated cooling effect under high temperature conditions is compared with the actual results in the validation dataset to verify the accuracy and generalization of the network.
[0059] Further optimization scheme, considering that in the current cooling effect prediction method, the main result achieved is to use neural network to establish the mapping between working condition data and cooling effect, which cannot be applied to the process of predicting high temperature working condition cooling effect by using low temperature working condition cooling effect designed by modeling theory, especially in the background of high temperature radiation, this method of extracting the influence of radiation on air film cooling effect is lacking in current technology. Therefore, this embodiment provides a neural network correction method for the radiation error of turbine blade modeling theory, including the following steps:
[0060] (1) Based on the turbine blade modeling theory, a cooling effect data set under radiation is constructed. The data samples in the data set consist of cooling effects under low temperature conditions, cooling effects under high temperature conditions, and modeling condition parameters;
[0061] (2) According to the size of the cooling effect data and the working condition data, a conditional generative adversarial network is constructed. The network mainly consists of a generator and a discriminator. The generator is U-Net, and the stability of the generator is improved by adding a residual network structure. Its input is the cooling effect data under low temperature conditions and the simulated working condition data, and its output is the cooling effect data under high temperature conditions. The discriminator is a binary classifier, which is trained in conjunction with the generator. Its input is the cooling effect data, and its output is the attribute of the cooling effect, that is, whether the cooling effect is a real result or generated by the generator. The network training data set is constructed using the data set established in (1) to train the network;
[0062] (3) Use the dataset established in (1) to construct a network training dataset and a verification dataset, and train and verify the conditional generative adversarial network constructed in (2).
[0063] Furthermore, in step 1 described in this embodiment, cooling efficiency training data is obtained through CFD numerical calculation, and the calculation model is a turbine blade film cooling unit. 210 sets of modeling processes are formulated, and the constructed modeling process is modeling from low-temperature conditions to high-temperature conditions, that is, the gas temperatures of the two conditions are different, and the remaining operating parameters (such as blowing ratio, momentum ratio, model size) need to meet the modeling theory requirements. And a medium gas temperature can be used as both a low temperature in one set of modeling processes and a high temperature in another set of modeling conditions, that is, the mutual combination of gas temperatures can construct more modeling processes, which can reduce the required CFD calculation conditions. In this example, a total of 140 sets of CFD calculation conditions are involved. The selected variables are: wall emissivity, absorption coefficient, and gas temperature. The CFD calculation result is the temperature field distribution of the cooling unit, which needs to be converted into cooling efficiency data when constructing the data set. The calculation method is as follows:
[0064]
[0065] Among them, T g is the gas temperature, T aw is the wall temperature under adiabatic conditions, T c The air conditioning temperature.
[0066] Furthermore, the reason why the generator in step 2 of this embodiment adopts U-Net is that the network structure has a strong feature extraction capability, can reduce the amount of required training data, and further improves the training efficiency of the network by adding a residual network structure. The discriminator promotes the optimization of the generator and improves the accuracy of the generated results by identifying the generated results of the generator until the discriminator cannot identify whether the generated results are generated, and a Nash equilibrium is reached. The L1 regularization term is added to the objective function used in the constructed conditional generative adversarial network, which is expressed as:
[0067]
[0068] Among them, G is the generator, D is the discriminator, x is the operating condition data, y is the cooling effect under high temperature conditions, and λ is the regularization factor.
[0069] Furthermore, the verification data set constructed in step 3 of the present embodiment has no intersection with the training data set, and at this time only the generator in the network needs to be used. The cooling effect under low temperature conditions and the working conditions of the simulation process in the verification set are input into the generator in turn, and the generated cooling effect under high temperature conditions is compared with the actual results in the verification set to verify the accuracy and generalization of the network.
[0070] The present invention quickly corrects the turbine blade modeling theory based on the conditional generative adversarial network, significantly reduces the prediction error of the cooling effect under high-temperature conditions, improves the applicability of the modeling theory, improves the temperature assessment accuracy of the engine turbine blades, and greatly improves the calculation efficiency, which has broad application prospects.
[0071] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A neural network correction method for the theoretical radiation error of turbine blade modeling, characterized in that: include: Constructing a turbine blade cooling data set, building a conditional generative adversarial network according to the sizes of cooling data and operating condition data in the turbine blade cooling data set, and training the conditional generative adversarial network to obtain a target conditional generative adversarial network; Based on the target condition generative adversarial network, the cooling effect of high temperature conditions is predicted according to the cooling effect of low temperature conditions and the operating parameters of the simulation process, and the accuracy and generalization of the target condition generative adversarial network are verified.
2. The neural network correction method for the theoretical radiation error of turbine blade modeling according to claim 1 is characterized in that: The process of building a turbine blade cooling effect dataset includes: Based on the turbine blade modeling theory, a dataset of turbine blade cooling effect under radiation is constructed; The data samples in the turbine blade cooling effect data set include cooling effect under low temperature conditions, cooling effect under high temperature conditions and simulation condition parameters.
3. The neural network correction method for the theoretical radiation error of turbine blade modeling according to claim 2 is characterized in that: Based on the turbine blade modeling theory, the process of constructing a turbine blade cooling effect dataset under radiation includes: The cooling effect training data is obtained by calculating the CFD numerical value through the calculation model of the turbine blade film cooling unit, and a modeling process is constructed based on the cooling effect training data to obtain the turbine blade cooling effect data set.
4. The neural network correction method for the theoretical radiation error of turbine blade modeling according to claim 3 is characterized in that: The process of constructing a modeling process based on the cold effect training data includes: Two working conditions with different gas temperatures are modeled from low temperature to high temperature, and the medium gas temperature is used as both the low temperature in one set of modeling processes and the high temperature in another set of modeling conditions. The modeling process is constructed by the mutual combination of gas temperatures.
5. The neural network correction method for turbine blade modeling theoretical radiation error according to claim 3 is characterized in that: The calculation of CFD values involves CFD calculation conditions; The variables selected for the CFD calculation conditions include: wall emissivity, absorption coefficient, and gas temperature.
6. The neural network correction method for turbine blade modeling theoretical radiation error according to claim 3 is characterized in that: The CFD numerical calculation result is the temperature field distribution of the cooling unit. When constructing the turbine blade cooling effect data set, the temperature field distribution is converted into cooling effect data. The formula expression is: Among them, T g is the gas temperature, T aw is the wall temperature under adiabatic conditions, T c The air conditioning temperature.
7. The neural network correction method for turbine blade modeling theoretical radiation error according to claim 1 is characterized in that: The L1 regularization term is added to the objective function used to construct the conditional generative adversarial network, which is expressed as: Among them, G is the generator, D is the discriminator, x is the operating condition data, y is the cooling effect under high temperature conditions, and λ is the regularization factor.
8. The neural network correction method for turbine blade modeling theoretical radiation error according to claim 1 is characterized in that: The conditional generative adversarial network includes a generator and a discriminator; The generator is a U-Net with a residual network structure added; The discriminator is a binary classifier, which is trained in conjunction with the generator. The input of the discriminator is cold effect data, and the output is the attribute of the cold effect.
9. The neural network correction method for turbine blade modeling theoretical radiation error according to claim 1 is characterized in that: The input of the target condition generation adversarial network is the cooling efficiency data under low temperature conditions and the simulation condition data, and the output is the cooling efficiency data under high temperature conditions.
10. The neural network correction method for turbine blade modeling theoretical radiation error according to claim 1, characterized in that: The process of verifying the accuracy and generalization of the target-conditional generative adversarial network includes: The turbine blade cooling effect data set is divided into a verification data set and a training data set. The cooling effect under low-temperature conditions and the working conditions of the simulation process in the verification data set are input into the generator in turn, and the generated cooling effect under high-temperature conditions is compared with the actual results in the verification data set to verify the accuracy and generalization of the network.