Gas turbine turbine blade root wheel groove profile optimization method based on info-GAN
By using an info-GAN generative adversarial network model combined with optimization algorithms, the problem of relying on engineering experience for the optimization design of turbine blade root grooves was solved. This enabled efficient and automated optimization, improved the structural strength and reliability of the blade root grooves, and made the technology applicable to a variety of turbine machinery.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XI AN JIAOTONG UNIV
- Filing Date
- 2023-12-06
- Publication Date
- 2026-07-31
AI Technical Summary
In the existing technology, the optimization design of turbine blade root grooves in gas turbines relies on engineering experience, which is costly to calculate and has a long optimization cycle. There is a lack of automated and efficient optimization methods.
By employing an info-GAN generative adversarial network model and combining it with optimization algorithms, a profile model is established for the root groove region of the gas turbine blade, enabling efficient and automated optimization design.
It enables rapid and accurate optimization of the root groove profile of gas turbine blades, improving the strength and reliability of the structure, and is applicable to various turbine machinery fields.
Smart Images
Figure CN117634084B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of turbine blade root groove profile optimization, specifically involving a gas turbine turbine blade root groove profile optimization method based on info-GAN. Background Technology
[0002] Blades are core components of turbine machinery such as gas turbines and aero engines, operating in complex environments for extended periods and are among the most vulnerable parts. The blade root-rotor groove structure, as the connection between the turbine blades and the rotor disk, is the primary load-bearing component for centrifugal forces during blade rotation. Therefore, the safety and reliability of the blade root-rotor groove structure are crucial for the normal and stable operation of the gas turbine. Thus, optimizing the design of the blade root-rotor groove structure to improve its strength and reliability is of significant engineering importance for ensuring the safe and stable operation of gas turbine units.
[0003] However, due to the complex structure and numerous geometric control parameters of the blade root groove region in gas turbines, and the high computational cost of optimizing and calculating this region, research on the optimization design of gas turbine blade root grooves is currently extremely limited. Previous optimization methods often rely on engineering experience and require significant manual intervention, resulting in long optimization cycles and considerable limitations in the optimization design. Summary of the Invention
[0004] The purpose of this invention is to provide a gas turbine blade root groove profile optimization method based on info-GAN in order to achieve efficient and automated optimization of gas turbine blade root grooves.
[0005] This invention is achieved using the following technical solution:
[0006] A gas turbine blade root groove profile optimization method based on info-GAN includes:
[0007] This method establishes a profile model of the turbine blade root groove region and uses the info-GAN method to build a generative adversarial network based on the profile image to predict the performance parameters of different blade root groove profile shapes. Based on this profile model and combined with optimization algorithms, an info-GAN-based optimization model for the turbine blade root groove profile is established, realizing efficient and automated optimization of the turbine blade root groove.
[0008] A further improvement of this invention is that the specific implementation method is as follows:
[0009] 1) Parametric model building of the root groove profile of the gas turbine blade
[0010] For a complete gas turbine blade, the blade root groove is selected, and its profile is extracted. Parametric modeling is used to reconstruct the profile, employing n profile geometric parameters R. i Establish a parameterized model of the root groove profile of a gas turbine blade, where i = 1, 2, 3, ..., n; after reconstruction, constrain the range of values and interrelationships of each profile geometric parameter according to the actual gas turbine blade conditions.
[0011] 2) Sample sampling, model reconstruction and calculation
[0012] For the parameterized model of the gas turbine blade root groove profile in step 1), the Latin hypercube sampling method is used to sample the n profile geometric parameters R. i Under the constraints of mutual relationships and value ranges, random sampling is performed to generate N samples, using K... j The expression is given, where j = 1, 2, 3, ..., N; the overall gas turbine blade is reconstructed by using the profile parameters of the blade root groove generated by random sampling, and finite element analysis is performed to obtain important stress field data affecting the safety and stability of the blade root groove;
[0013] 3) Organize the data and create a dataset.
[0014] Based on the important stress field data obtained in step 2), a performance evaluation index P for the root groove of the gas turbine blade is established, and a dataset corresponding to the root groove profile image of the gas turbine blade and the performance evaluation index P is established. The dataset is divided into training and validation parts for use in training the model.
[0015] 4) Build and train the info-GAN generative adversarial network model
[0016] Using the dataset obtained in step 3), train the info-GAN generative adversarial network model to achieve processing of arbitrary geometric parameters R. i The generated gas turbine blade root groove profile is used to accurately predict its blade root groove performance evaluation index P.
[0017] 5) Optimize using optimization algorithms.
[0018] The performance evaluation index P of the blade root groove is set as the optimization objective. Combined with the actual situation of the gas turbine blade, the range of geometric parameters in step 2) is set as the constraint condition. New gas turbine blade root groove profile images are continuously generated, and the evaluation index P value is returned by the info-GAN generative adversarial network model. Then, the optimization algorithm is used to find the best value. The process ends when the set optimization objective is met or the number of iterations reaches the maximum.
[0019] A further improvement of the present invention is that step 1) is implemented as follows:
[0020] 101) Perform CAE strength simulation analysis on the complete gas turbine blade-groove model to obtain its stress and strain physical field distribution, and analyze the stress and strain hazards and concentration areas in the blade root groove region under working load and working temperature conditions.
[0021] 102) Extract the blade root groove profile of the complete turbine blade-groove model to be optimized separately, prioritizing the stress-strain critical areas, and confirm the n profile geometric variable parameters R. i In addition to several fixed parameters, a parameterized model of the turbine root groove profile of a gas turbine can be completely constructed using n profile geometric variable parameters and several fixed parameters, and the n profile geometric variable parameters are independent of each other and can basically cover all stress and strain dangerous areas of the blade root groove.
[0022] 103) Based on the actual conditions of the gas turbine blades and the acceptable optimization range, determine the upper and lower bounds of different parameters, as well as the constraint region of the blade root groove. The final gas turbine blade root groove parameters are expressed in the following form:
[0023] R TB =[R1,R2,R3,...,R n ]
[0024] D i,min ≤R i ≤D i,max
[0025] F(R TB )∈A
[0026] Where: R TB D represents the geometric variable parameter matrix of the blade root groove. i，min and D i，max Each variable parameter R represents a different parameter. i The lower and upper bounds of the optimizable range are given by A, where A represents the constraints and range of the blade root groove.
[0027] A further improvement of the present invention is that step 2) is implemented as follows:
[0028] 201) The geometric variable parameter matrix R of the blade root groove established in step 1). TB Using the Latin hypercube sampling method, the geometric variable parameters R of the n-shaped profile are... i upper and lower bounds D i，min and D i，max Within the range, perform random sampling to select N R values. TB sample;
[0029] 202) Using N samples, the code automatically draws the corresponding two-dimensional profile diagrams of the root grooves of the gas turbine blades, and generates the corresponding three-dimensional numerical analysis calculation models in batches through the self-developed command flow of the CAE simulation software, and performs batch strength simulation calculations on the corresponding models of the N gas turbine blade root grooves.
[0030] 203) Batch store the stress field results obtained from the intensity simulation and correspond them one-to-one with the profile image of each blade root wheel groove for subsequent data processing.
[0031] A further improvement of the present invention is that step 3) is implemented as follows:
[0032] 301) Based on the stress field results calculated in step 2), and according to the stress maximum and concentration regions obtained from the simulation of the original model in step 1), identify the maximum stress values of the m stress-rich regions as the characteristic stress values of the stress field, and assign them the values σ1, σ2, σ3...σ m This means that, based on the magnitude of the relative stress value and the importance of its location, a weight k1, k2, k3...k is assigned to each characteristic stress value. m And establish the performance evaluation index P of the turbine blade root groove of the gas turbine, as shown in the following formula:
[0033]
[0034] Among them, P origin k is the performance evaluation index of the original model. i,origin , σ i,origin It consists of each feature stress value of the original model and its corresponding weight;
[0035] 302) Based on the operation 202) in step 2), create a two-dimensional profile for N samples and assign the corresponding blade root groove performance evaluation index P to the profile as the label for each sample; organize the images and labels of the N samples into the input format specified by the generative adversarial network model, and divide them into training set and validation set according to the specified ratio.
[0036] A further improvement of this invention is that step 4) is implemented as follows:
[0037] The Info-GAN generative adversarial network model evolved from the GAN model and consists of a generator G, a classifier Q, and a discriminator D. Both the generator G and the discriminator D are composed of independent convolutional neural networks. The input to the generator G consists of random noise z and a hidden code c. The hidden code c is used by the convolutional neural network to predict and output a generated sample G(z), which is then fed into the discriminator D along with a real sample x. The discriminator D judges whether the sample is real or fake. The discriminator D and the classifier Q use the same convolutional neural network. The discriminator D outputs the judgment of whether the sample is real or fake, while the classifier Q uses the convolutional neural network to predict the hidden code c of the generated sample G(z) and outputs the predicted value c'. If the hidden code c input to the generator has a clear impact on the generated sample, i.e., the two are highly correlated, the output c' of Q should be as consistent as possible with the hidden code c. To improve the correlation between the hidden code c and the generated sample, and to ensure that a certain dimension of the hidden code corresponds to certain semantic information of the generated sample, Info-GAN maximizes the mutual information between the two.
[0038] InfoGAN incorporates mutual information into its loss function, adding a regularization term λI(c,G(z,c)) to the original GAN's loss function, where λ is a hyperparameter. A regularization function for the original loss function is proposed, and the training objective function is shown below:
[0039]
[0040] in: This represents the objective function of the GAN model;
[0041] The goal of generator G is to generate fake samples that can fool discriminator D. The loss function is shown in the following equation:
[0042]
[0043] The discriminator D has the same goal as the original GAN: to judge whether a sample is real or fake. It identifies real samples as real and fake samples as fake. The loss function is shown below:
[0044]
[0045] For the joint network of generator G and auxiliary network Q, in order to make the hidden code c and the output G(z,c) highly correlated, the loss function is as follows:
[0046] L info =maxL I (G,Q)=maxI(c;G(z,c))
[0047] Specifically, firstly, the matrix data of the two-dimensional black-and-white pixel image of the blade root groove profile is used as the input of the hidden code c. The predicted blade root groove performance evaluation index P is obtained through the convolutional neural network of the generator G, serving as the generated sample G(z). The simulated P value of the blade root groove profile is then used as the real sample x and fed into the discriminator D for judgment and prediction. The prediction returns prediction data that closely approximates the real blade root groove profile image matrix. After repeated training and verification, the system finally achieves the desired performance for a given random condition satisfying R... TB The parameterized blade root groove profile is used to return a predicted value P' that is sufficiently close to the P value obtained from the simulation calculation of the corresponding real model.
[0048] A further improvement of the present invention is that step 5) is implemented as follows:
[0049] 501) The geometric variable parameter matrix R of the blade root groove TB As optimization variable parameters, within the optimizable lower and upper bounds of each optimization variable, a genetic algorithm is selected to establish an optimization calculation model for optimization; through the code model established in step 2), the corresponding two-dimensional profile diagram of the gas turbine blade root groove is automatically drawn, and the profile diagram is imported into the generative adversarial network model trained in step 4). The blade root groove performance evaluation index P value returned by the generative adversarial network is imported back into the target parameters of the optimization algorithm for optimization calculation.
[0050] 502) By optimizing the algorithm until convergence, the blade root groove profile corresponding to the best performance evaluation index P value of the blade root groove is reconstructed to obtain the optimized gas turbine blade model.
[0051] A further improvement of the present invention is that, in step 5), the optimization algorithm adopts a pattern search or a genetic algorithm.
[0052] The present invention has at least the following beneficial technical effects:
[0053] Compared to traditional methods, this invention provides an info-GAN-based optimization method for the root groove profile of gas turbine blades. This method can optimize the root groove profile of a given gas turbine blade, starting from the actual blade model and calculating the root groove profile. The optimized profile is then returned, and the blade can be further reconstructed. This allows for accurate, flexible, and rapid acquisition of optimized gas turbine blades that meet engineering requirements.
[0054] Furthermore, in the generative model of infoGAN, the present invention uses real labels as part of the features and inputs them into the generative and discriminative networks for training, providing better guidance for generative and discriminative behaviors. This can solve the shortcomings of conventional generative adversarial networks, such as training difficulties and high sensitivity to hyperparameters.
[0055] Furthermore, this invention uses the value range of the real label as a constraint condition and employs methods including but not limited to genetic algorithms, pattern search algorithms, and various biomimetic optimization algorithms to perform optimization, thereby obtaining an optimized blade root groove shape within the set value range of the real label.
[0056] Furthermore, this invention employs a method of directly optimizing the blade root groove profile region, which greatly improves the applicability of the invention. All blade root groove profiles obtained during the optimization process of this invention, after being scaled proportionally to a specified size, are not limited to the original gas turbine blades, but can be applied to any gas turbine blade, and even to a wider range of turbine machinery fields such as heavy-duty steam turbines and aero engines.
[0057] Furthermore, during the implementation of this invention, a large number of gas turbine blade root groove profile diagrams and their corresponding blade root groove performance evaluation index P will be obtained, which can build a complete gas turbine blade root groove profile database, providing important data and styling references for the field of gas turbine blade design.
[0058] Furthermore, the generative adversarial network model established in this invention will accumulate a large amount of raw data during multiple implementations. Its function is not limited to achieving optimization, but can also achieve the rating and prediction of the quality of the root groove shape of any target gas turbine blade.
[0059] In summary, this invention provides a highly efficient, stable, and significantly effective method for optimizing the root groove profile of gas turbine blades. It is widely applicable to the design optimization of root grooves for various turbine blades. Attached Figure Description
[0060] Figure 1 This is a flowchart of a gas turbine blade root groove profile optimization method based on info-GAN according to the present invention.
[0061] Figure 2 This is a schematic diagram of the info-GAN model structure in this invention.
[0062] Figure 3 This is a schematic diagram illustrating the simplified process and parameter settings of the root groove profile of the gas turbine blade in an embodiment of the present invention.
[0063] Figure 4 This is a flowchart illustrating an embodiment of the present invention. Detailed Implementation
[0064] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.
[0065] like Figure 1 As shown, the present invention provides a method for optimizing the root groove profile of gas turbine blades based on info-GAN, which mainly consists of 6 steps, which will be described below in conjunction with... Figure 1 and Figure 4 The implementation method of the embodiment is explained step by step.
[0066] 1. Parametric model construction of the root groove profile of the gas turbine blade:
[0067] For a specific gas turbine blade, a parametric model is built, such as... Figure 3 As shown, the gas turbine blade shape is first determined, and the profile of the blade root groove region is extracted through a certain section and simplified. After simplification, the blade root region profile is parametrically modeled, and the groove region is set to follow the blade root profile while ensuring basic geometric characteristics remain consistent. Furthermore, CAE simulation calculations are used to identify the stress-critical locations in the blade root groove region, and a parametric model that basically covers these critical locations is established.
[0068] The parameter settings in this embodiment are as follows: Figure 3 As shown, where: R1 represents the radius of the first arc at the leaf root; θ1 represents the central angle of the first arc at the leaf root; L1 represents the width of the contact surface of the first pair of teeth; R2, L2, and R3 represent the dimensions of the protruding tooth portions; L4 represents the width of the contact surface of the second to fourth pairs of teeth. D1, D2, D3, and D4 represent the throat widths of the first to fourth pairs of teeth, respectively; H1, H2, and H3 represent the vertical heights of the first to second pairs of teeth, the second to third pairs of teeth, and the third to fourth pairs of teeth, respectively. Considering the independence of variables, R2, L2, and R3 are treated as dependent variables. Considering the consistency of the leaf shape, the values of H2 and H3 are taken as fixed multiples of H1. Therefore, a total of 9 independent geometric parameters are determined, forming the following parameterization matrix:
[0069] R TB =[R1,θ1,L1,L4,D1,D2,D3,D4,H1]
[0070] Furthermore, for each geometric feature, considering the specific dimensions of the model, its upper and lower bounds for optimization are determined individually, and overall constraints are established. For example, the maximum width of the first tooth at the blade root should be less than the width of the blade's intermediate body, and the overall range of the blade root and wheel groove should be within the range of a single sector of the wheel disk. The final determined parameterization matrix and range are as follows:
[0071] R TB =[R1,θ1,L1,L4,D1,D2,D3,D4,H1]
[0072] R TB，min ≤R TB ≤R TB，max
[0073] F(R TB )∈A
[0074] 2. Sample Sampling, Model Reconstruction and Calculation
[0075] Based on the above parameterized model, the initial parameterized matrix R is obtained for the original model. TB,origin Furthermore, a number of samples were obtained by randomly sampling the nine variables in the parameterized matrix using the Latin hypercube sampling method. For each sample, a model reconstruction was performed. Considering both computational economy and accuracy, a two-dimensional model or a locally simplified model of the blade root and wheel groove could be used for numerical analysis. Finally, the stress field distribution corresponding to the profile of each sample was obtained.
[0076] Based on the calculated stress field results and the simulated areas of maximum and concentrated stress, the maximum stress values in m stress-rich regions are identified as the characteristic stress values of the stress field, denoted by σ1, σ2, σ3...σ m This means that, based on the magnitude of the relative stress value and the importance of its location, a weight k1, k2, k3...k is assigned to each characteristic stress value. m And establish the performance evaluation index P of the turbine blade root groove of the gas turbine, as shown in the following formula:
[0077]
[0078] Among them, P origin k is the performance evaluation index of the original model. i,origin , σ i,origin It represents each characteristic stress value of the original model and its corresponding weight.
[0079] 3. Organize the data and create a dataset.
[0080] like Figure 4 As shown, the data is organized, the profile of each sample is uniformly processed, and the calculated performance evaluation index P is attached as a label to the profile of each sample.
[0081] 4. Build and train the info-GAN generative adversarial network model
[0082] like Figure 2 As shown, an info-GAN generative adversarial network model is constructed, and pre-processed sample sets (training and validation sets) are imported into the info-GAN generative adversarial network model for training and validation.
[0083] 5. Optimize using optimization algorithms.
[0084] The geometric variable parameter matrix R of the blade root groove TB As optimization variables, within the optimizable lower and upper bounds of each variable, a genetic algorithm is selected to establish an optimization computation model for optimization. The genetic algorithm flowchart is as follows: Figure 4 The flowchart is shown. Genetic Algorithm (GA) is a biomimetic optimization algorithm that utilizes biological inheritance and is widely used in many fields. The flowchart of the genetic algorithm is shown below. Figure 4 The flowchart shows operations including population initialization, fitness calculation, selection, crossover, and mutation. In this embodiment, the maximum number of generations is set to 100, the crossover probability to 0.8, the mutation probability to 0.2, and the population size to 300.
[0085] Using the code model established in step 2, the corresponding two-dimensional profile of the gas turbine blade root groove is automatically drawn, and the profile is imported into the generative adversarial network model trained in step 4. The blade root groove performance evaluation index P value returned by the generative adversarial network is imported back into the target parameters of the optimization algorithm for optimization calculation.
[0086] 6. Complete optimization and reconstruct the blade model.
[0087] Once the optimization algorithm satisfies the optimization objective and converges, the final optimal parameter matrix R can be obtained. TB,opt The two-dimensional profile of the blade root groove region is obtained by reconstructing the parameter matrix, and then the profile is applied to the original model to reconstruct the blade root groove region of the original blade. After reconstruction, simulation analysis and comparison of the models before and after optimization can be performed, and experimental verification designs can be added as appropriate to ensure the optimization effect.
[0088] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.
Claims
1. A method for optimizing the root groove profile of a gas turbine blade based on info-GAN, characterized in that, include: This method establishes a profile model of the turbine blade root groove region and uses the info-GAN method to build a generative adversarial network based on the profile image to predict the performance parameters of different blade root groove profile shapes. Based on this profile model and combined with optimization algorithms, an info-GAN-based optimization model for the turbine blade root groove profile is established to optimize the turbine blade root groove. The specific implementation method is as follows: 1) Parametric model building of the root groove profile of the gas turbine blade For a complete gas turbine blade, the blade root groove is selected, and its profile is extracted. Parametric modeling is used to reconstruct the profile, using a total of [number missing] [units missing]. n Geometric parameters of individual shapes R i A parametric model of the root groove profile of a gas turbine blade is established, in which... i =1, 2, 3..., n After reconstruction, the range of values and interrelationships of each profile geometric parameter are constrained according to the actual condition of the gas turbine blades. 2) Sample sampling, model reconstruction and calculation For the parameterized model of the gas turbine blade root groove profile in step 1), the Latin hypercube sampling method is used to... n Geometric parameters of individual shapes R i Under the constraints of mutual relationships and value ranges, random sampling is performed to generate... N One sample, using K j It means that, among them, j =1, 2, 3..., N By using the profile parameters of the blade root groove generated by random sampling, the turbine blade of the entire gas turbine is reconstructed and finite element analysis is performed to obtain important stress field data affecting the safety and stability of the blade root groove. 3) Organize the data and create a dataset. Based on the key stress field data obtained in step 2), a performance evaluation index P for the root groove of the gas turbine blade is established. A dataset corresponding to the gas turbine blade root groove profile image and the performance evaluation index P (i.e., the label) is also established. This dataset is divided into training and validation parts for use in training the model, including: 301) Based on the stress field results calculated in step 2), and according to the stress maximum and concentration regions obtained from the original model simulation in step 1), identify the maximum stress values of m stress-rich regions as the characteristic stress values of the stress field, and respectively use... This means that a weight is assigned to each characteristic stress value based on the magnitude of its relative stress value and the importance of its location. And establish a performance evaluation index for the root groove of gas turbine blades. P As shown in the following formula: in, It is the performance evaluation index of the original model. , It consists of each feature stress value of the original model and its corresponding weight; 302), Based on operation 202) in step 2), give N Create a two-dimensional profile diagram for each sample and assign the corresponding blade root groove performance evaluation index to the profile diagram. P , as the label for each sample; N The images and labels of each sample are organized into the input format specified by the generative adversarial network model and divided into training set and validation set according to the specified ratio; 4) Build and train the info-GAN generative adversarial network model Using the dataset obtained in step 3), train the info-GAN generative adversarial network model to achieve processing of arbitrary geometric parameters. R i The generated gas turbine blade root groove profile accurately predicts its blade root groove performance evaluation index. P ; 5) Optimize using optimization algorithms Setting the performance evaluation index of the blade root groove P To optimize the objective, and considering the actual conditions of the gas turbine blades, the range of geometric parameter values in step 2) is set as a constraint to continuously generate new gas turbine blade root groove profile images. An evaluation index is then returned using an info-GAN generative adversarial network model. P The value is then used to find the optimal value using an optimization algorithm; the process ends when the set optimization objective is met or the number of iterations reaches its maximum.
2. The method for optimizing the root groove profile of a gas turbine blade based on info-GAN according to claim 1, characterized in that, The specific implementation method of step 1) is as follows: 101) Perform CAE strength simulation analysis on the complete gas turbine blade-groove model to obtain its stress and strain physical field distribution, and analyze the stress and strain hazards and concentration areas in the blade root groove region under working load and working temperature conditions. 102) Extract the blade root groove profile of the complete gas turbine blade-groove model to be optimized separately, prioritizing areas with critical stress and strain, and confirm... n Geometric variable parameters of individual shape lines R i And several fixed parameters, so that by n A complete parametric model of the turbine blade root groove profile can be constructed using one set of geometric variable parameters and several fixed parameters, and this ensures... n The geometric parameters of each profile are independent of each other and can basically cover the stress and strain critical areas of all blade root grooves. 103) Based on the actual conditions of the gas turbine blades and the acceptable optimization range, determine the upper and lower bounds of different parameters, as well as the constraint region of the blade root groove. The final gas turbine blade root groove parameters are expressed in the following form: in: R TB This represents the geometric variable parameter matrix of the blade root groove. D i，min and D i，max Each variable parameter represents a different parameter. R i The lower and upper bounds of the optimizable range, This indicates the constraints and range of the blade root groove.
3. The method for optimizing the root groove profile of a gas turbine blade based on info-GAN according to claim 2, characterized in that, The specific implementation method of step 2) is as follows: 201) The geometric variable parameter matrix of the blade root groove established in step 1). R TB By using the Latin hypercube sampling method, in n Geometric variable parameters of individual shape lines R i upper and lower boundaries D i，min and D i，max Within the range, perform random sampling to select N items. R TB sample; 202) Using N samples, the code automatically draws the corresponding two-dimensional profile diagrams of the root grooves of the gas turbine blades, and generates the corresponding three-dimensional numerical analysis calculation models in batches through the self-developed command flow of the CAE simulation software, and performs batch strength simulation calculations on the corresponding models of the N gas turbine blade root grooves. 203) Batch store the stress field results obtained from the intensity simulation and correspond them one-to-one with the profile image of each blade root wheel groove for subsequent data processing.
4. The method for optimizing the root groove profile of a gas turbine blade based on info-GAN according to claim 3, characterized in that, The specific implementation method of step 4) is as follows: The Info-GAN generative adversarial network model evolved from the GAN model and consists of a generator G, a classifier Q, and a discriminator D. Both the generator G and the discriminator D are composed of independent convolutional neural networks. The input to the generator G consists of random noise z and a hidden code c. The hidden code c is used by the convolutional neural network to predict and generate a sample G(z), which is then fed into the discriminator D along with a real sample x. The discriminator D judges whether the sample is real or fake. The discriminator D and the classifier Q use the same convolutional neural network. The discriminator D outputs the judgment of whether the sample is real or fake, while the classifier Q uses the convolutional neural network to predict the hidden code c of the generated sample G(z) and outputs the predicted value c'. If the hidden code c input to the generator has a high correlation with the generated sample, the output c' of Q will be consistent with the hidden code c. Info-GAN maximizes the mutual information between the hidden code c and the generated sample. InfoGAN incorporates mutual information into its loss function, adding a regularization term to the original GAN's loss function. ,in These are hyperparameters. A regularization function for the original loss function is proposed. The training objective function is shown below: in: , represents the objective function of the GAN model; The goal of generator G is to generate fake samples that can fool discriminator D. The loss function is shown in the following equation: The discriminator D has the same goal as the original GAN: to judge whether a sample is real or fake. It identifies real samples as real and fake samples as fake. The loss function is shown below: For generators G and auxiliary network Q The joint network, in order to enable implicit coding c and output G(z,c) The correlation is strong, and the loss function is as follows: Specifically, the matrix data of the two-dimensional black-and-white pixel image of the blade root groove profile is first used as the input of the hidden code c, and the predicted blade root groove performance evaluation index is obtained through the convolutional neural network of the generator G. P The value is used as the generated sample G(z), and the simulation calculation of the blade root groove profile is used. P The value is used as a real sample x and is fed into the discriminator D for judgment and prediction. The prediction will return predicted data that is close to the real blade root groove profile image matrix. After repeated training and verification, it is finally possible to achieve the prediction of a random given value that satisfies the following conditions. R TB The parameterized blade root groove profile is returned as a value sufficiently close to the simulation result of its corresponding real model. P Predicted value P' .
5. The method for optimizing the root groove profile of a gas turbine blade based on info-GAN according to claim 4, characterized in that, The specific implementation method of step 5) is as follows: 501) The geometric variable parameter matrix of the blade root groove R TB As optimization variables, within the optimizable lower and upper bounds of each optimization variable, a genetic algorithm is selected to establish an optimization calculation model for optimization. Using the code model established in step 2), the corresponding two-dimensional profile of the gas turbine blade root groove is automatically drawn, and the profile is imported into the generative adversarial network model trained in step 4). The performance evaluation index of the blade root groove returned by the generative adversarial network is then used. P The value is substituted back into the target parameters of the optimization algorithm for optimization calculations. 502) Calculate the performance evaluation index of the blade root groove through optimization algorithm until convergence. P By reconstructing the model using the blade root groove profile corresponding to the optimal value, the optimized gas turbine blade model can be obtained.
6. The method for optimizing the root groove profile of a gas turbine blade based on info-GAN according to claim 1, characterized in that, In step 5), the optimization algorithm uses a genetic algorithm.