Turbine blade geometric structure uncertainty quantitative analysis system

Abstract

The invention discloses a turbine blade geometric structure uncertainty quantitative analysis system. The system comprises a CAD model construction module, an uncertainty sampling module, a CFD batch calculation module, a DNN model training module, an uncertainty analysis module and a turbine performance optimization module. According to the analysis system, on one hand, the efficiency and the stability of large-batch uncertainty design sample set CFD calculation are improved to a great extent, and application of an uncertainty optimization design technology in the aspect of actual engineering is facilitated; and on the other hand, by means of the system, the structure of the geometric model of the turbine blade is parameterized, sampling calculation of uncertainty design can be carried out on as many as dozens of parameters at the same time, relatively important parameters are screened on the basis of sensitivity analysis subsequently, the dimension reduction target is achieved, the problem of dimension disaster in traditional uncertainty analysis is solved, and the accuracy of the system is improved. And the accuracy and comprehensiveness of uncertainty analysis of the turbine blade are improved.

Description

Technical Field

[0001] The present invention relates to a system for quantifying the geometric structure uncertainty of turbine blades, belonging to the technical field of turbomachinery. Background Art

[0002] In an aeroengine, the turbine plays a key role due to its mission of energy conversion, and directly affects the aerodynamic performance and service life of the whole engine. Since the turbine operates under harsh conditions of high temperature, high pressure and high speed for a long time, temperature, thermal stress and ablation phenomena will seriously affect the operation of turbine blades. In recent years, with the increasing requirements for operating conditions, the turbine material itself cannot withstand such high-temperature incoming flow. Therefore, the cooling performance of the turbine becomes crucial. In a turbine, manufacturing geometric uncertainty directly affects the internal flow and heat transfer process of the turbine, thereby leading to the decline of aerothermal performance and reliability problems, resulting in a large dispersion in the service life of the hot-end components of the aeroengine. At the same time, with the continuous improvement of the requirements for turbine performance indicators and service life, the requirements for the refined design of the turbine are becoming more and more stringent. It is of great significance to determine and minimize the influence of various geometric uncertainties on performance as much as possible. In recent years, a series of research results have been achieved in turbine performance design based on different uncertainty quantification (UQ) methods.

[0003] Currently, the research on the uncertainty quantification analysis of turbine blades is mainly divided into two categories: one mainly focuses on the influence of key geometric parameters on the blade surface; the other focuses on the influence of geometric parameters of the cooling system. Most of the technical solutions in the research select a small number of geometric uncertainty parameters as inputs and some key performance indicators as outputs. After collecting uncertainty samples through numerical methods such as Latin Hypercube Sampling (LHS), the CFD calculations are manually carried out on the geometric models corresponding to the sample sets one by one to obtain the results. Finally, the calculation results are used as the training set and validation set to establish and train a surrogate model, and the uncertainty quantification analysis and multidisciplinary optimization of the selected geometric parameters are carried out.

[0004] Most of the existing research on the uncertainty quantification analysis of turbine blades only considers a few parameters. However, the turbomachinery itself is complex and there are many geometric uncertainty variables introduced during manufacturing and use. These variables not only affect the performance individually, but also the interaction between them will further cause changes in performance. Therefore, only considering a few parameters is not comprehensive and accurate enough. On the other hand, for the uncertainty quantification research on complex multi-variable problems, not only a large number of manual CFD calculations are required, resulting in a long design cycle, but also due to the high dimensionality of the parameter space, the "curse of dimensionality" will occur, and the calculation time and cost are unaffordable in engineering.

[0005] Therefore, the development of suitable high-dimensional dimensionality reduction methods, which can be applied to the actual engineering field, is of great significance. At the same time, since the turbine design objectives include multi-disciplinary and multi-objective constraints such as aerodynamics, heat transfer, and structure, the development of suitable multi-disciplinary and multi-objective optimization methods is crucial for turbine design. Summary of the Invention

[0006] In view of the problems existing in the above-mentioned background technology, the present invention provides a system for quantifying the geometric structure uncertainty of turbine blades, which can be used as an automated design platform for multi-disciplinary uncertainty analysis of turbine blades. The system of the present invention integrates multiple modules, and users do not need to manually call each module one by one or integrate various algorithms and software by themselves, and can easily complete the uncertainty quantification analysis work, which greatly improves the efficiency and stability of CFD calculations for a large number of uncertainty design sample sets, and helps to promote the application of uncertainty optimization design technology in actual engineering. On the other hand, by parameterizing the structure of the geometric model of the turbine blade with this system, sampling calculations for uncertainty design of up to dozens of parameters can be carried out simultaneously. Subsequently, based on sensitivity analysis, relatively important parameters are screened to achieve the dimensionality reduction goal, solving the problem of "curse of dimensionality" in traditional uncertainty analysis and improving the accuracy and comprehensiveness of uncertainty analysis of turbine blades.

[0007] The present invention is implemented by the following technical solutions:

[0008] A system for quantifying the geometric structure uncertainty of turbine blades includes a CAD model construction module, an uncertainty sampling module, a CFD batch calculation module, a DNN model training module, an uncertainty analysis module, and a turbine performance optimization module; the CAD model construction module is used to parameterize the structure of the geometric model of the turbine blade; the uncertainty sampling module is used to select geometric parameters according to the research objective, determine the uncertainty range of the geometric parameters according to the manufacturing and usage conditions, and perform sampling within the uncertainty range of the geometric parameters; the CFD batch calculation module is used to batch process the sampling results of the uncertainty sampling module, specifically including mesh generation, preprocessing, solution, and post-processing; the DNN model training module is used to train a DNN model according to the calculation results of the CFD batch calculation module, the input of the DNN model is geometric parameters, and the output is the target physical quantity; the uncertainty analysis module is used to perform sensitivity analysis on the geometric parameters selected by the uncertainty sampling module according to the target physical quantity and extract key geometric parameters; the turbine performance optimization module is used to perform deterministic optimization and uncertainty optimization on the key geometric parameters.

[0009] In the above technical solution, further, the geometric model of the turbine blade can be a model of a fully cooled turbine blade with a combined configuration of impingement and film cooling,... etc.

[0010] Furthermore, the CAD model construction module is used to parameterize the structure of the geometric model of the turbine blade. Specifically, the structure of the geometric model of the turbine blade includes: the curvature of the blade inner cavity; the blade thickness distribution; the blade tip shape; the blade tail shape; the aperture, angle, distribution gap, and number of film cooling holes; the aperture, angle, distribution gap, and number of impingement holes; the thickness, distribution gap, and number of rib walls; the radius, distribution gap, and number of turbulators.

[0011] Furthermore, the uncertainty sampling module can specifically adopt the Latin hypercube sampling method to randomly sample the selected geometric parameters.

[0012] Furthermore, the deterministic optimization is specifically to select a best set of key geometric parameters according to the target physical quantity; the uncertainty optimization is specifically to select a set of key geometric parameters according to the performance dispersion under the uncertainties brought by manufacturing and use, so as to maximize the robustness of the turbine system.

[0013] Furthermore, the mesh generation is specifically to generate a mesh for the geometric model of the turbine blade to discretize the solution domain; the preprocessing specifically includes defining the initial and boundary conditions and material properties; the solution is specifically to numerically solve the fluid equations; the postprocessing is to statistically analyze and visualize the numerical solution results.

[0014] Furthermore, the DNN model training module is used to train a DNN model based on the calculation results of the CFD batch calculation module. The specific method is as follows: randomly divide the calculation results of the CFD batch calculation module into K folds based on K-fold cross-validation, and use each fold as a validation set to train the DNN model in multiple rounds; optimize the hyperparameters of the DNN model hidden layer node combination based on the Grid Search method; extract the key geometric parameters based on the uncertainty analysis module, and use the key geometric parameters as the input of the new DNN model to continue training until the model accuracy reaches above the set threshold and the retained parameter dimensions meet the calculation requirements. The inner-loop dimensionality reduction method between the above DNN model and the uncertainty analysis module can screen out the key geometric parameters, and can solve the following problems: the problem that the neural network accuracy is insufficient due to too many selected parameters and the sparsity of the dataset, so that the sensitivity size relationship between two parameters that are close in the SHAP mean graph cannot be determined, and the problem of low calculation efficiency due to the complex geometric features of the turbine cooling blade.

[0015] Furthermore, the uncertainty analysis module is used to perform sensitivity analysis on the geometric parameters selected by the uncertainty sampling module based on the SHAP analysis method (specifically, the SHAP values of the geometric parameters can be calculated using the Kernel Explainer in the SHAP library of Python), and extract key geometric parameters. The specific method is as follows: calculate the SHAP values of the geometric parameters selected by the uncertainty sampling module. The larger the SHAP value of a geometric parameter, the greater its sensitivity. Select the geometric parameters with greater sensitivity as key geometric parameters according to the SHAP values. The specific calculation formula for the SHAP value of a geometric parameter is:

[0016]

[0017] δ i (S) = EY k (S ∪ (x i )) - EY k (S)

[0018] EY k (S) = E(Y(x k ))|S)

[0019] where, x i represents the i-th input feature vector, and the input feature vector is composed of geometric parameters; X is the set of all other input feature vectors that do not include x i . N represents the set of all feature vectors; Y(x k ) is the output vector corresponding to x k .

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] The present invention constructs a system for quantifying the geometric structure uncertainty of turbine blades, which integrates multiple modules. In terms of efficiency, the system performs a large number of uncertainty sampling calculations through an automated process and algorithms, eliminating the time and labor costs of manual sampling one by one and accelerating the calculation process through parallel computing technology, thereby improving the calculation efficiency. In terms of result processing, the system can generate visual results and reports, summarize the analysis results and data, facilitating subsequent decision-making and communication. In terms of repeatability and scalability, the system can repeatedly analyze the same problem by establishing a standardized uncertainty analysis process and tools, and can be quickly applied to other projects. In terms of integration and collaboration, the system can be integrated with existing engineering software and data systems to achieve automated data transmission and processing, support multi-user collaboration, and realize distributed uncertainty analysis and design. The structural parameters of the geometric model of the turbine blade are parameterized. In parametric design, these parameters can be adjusted to change the shape, size, and features of the object without reconstructing the entire geometric model. The parameterization of the geometric structure can improve the flexibility, efficiency, and accuracy of the design, accelerate the design process, reduce complexity, and make the design work more systematic and engineering-oriented. The Latin hypercube sampling method is used for uncertainty sampling calculations. By evenly dividing the sample space into continuous regions, a sample value is selected from each region to construct a sampling set. Compared with traditional sampling methods, the Latin hypercube sampling method can improve the sampling efficiency and increase the diversity of sampling points to better explore and represent the entire sample space. The surrogate model is trained based on the Grid Search method and the K-fold cross-validation method. Facing the problem of data sparsity in the training of the constructed deep neural network (DNN) model, the K-fold cross-validation method can make more full use of the data and reduce the bias of the model evaluation results. The dimensionality reduction is performed in an inner loop based on the SHAP analysis method. In traditional black-box models such as deep neural networks, they usually have a large number of parameters and complex computational structures, making it difficult to understand and explain their internal operating mechanisms. The selected SHAP (Shapley Additive Explanations) value analysis method can well explain the model and provide the sensitivity of individual geometric variables. On the other hand, this inner-loop dimensionality reduction method solves the contradiction of insufficient neural network accuracy caused by a large number of selected parameters and sparse data sets, thereby obtaining the geometric features that have the greatest impact on the turbine cooling blades and greatly improving the calculation efficiency. The present invention synthesizes the interactions of multiple disciplines (such as aerodynamics, heat transfer, and structure), simultaneously considers the design variables and uncertainties in the system, and optimizes the target physical quantity. Multidisciplinary optimization simplifies the coupling relationship between design variables and performance indicators through machine learning methods by establishing collaborative relationships between different disciplines, and comprehensively considers to obtain the optimal design. Description of the Drawings

[0022] Figure 1It is the workflow diagram of the system for quantifying the geometric structure uncertainty of the turbine blade of the present invention.

[0023] Figure 2 It is the parametric model of the turbine cooling blade.

[0024] Figure 3 It is the SHAP mean value diagram (left) and the SHAP value summary diagram (right) of 15 groups of parameters.

[0025] Figure 4 It is the flowchart of the dimensionality reduction and optimization neural network.

[0026] Figure 5 It is the R 2 variation diagram during the dimensionality reduction process.

[0027] Figure 6 It is the Pareto point diagram in the sample space.

[0028] Figure 7 It is the Pareto curve diagram fitted based on the genetic algorithm. Specific implementation manners

[0029] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0030] As Figure 1 shown, a system for quantifying the geometric structure uncertainty of a turbine blade of the present invention integrates multiple modules, specifically including a CAD model construction module, an uncertainty sampling module, a CFD batch calculation module, a DNN model training module, an uncertainty analysis module, and a turbine performance optimization module.

[0031] The present invention includes two major functions. The first part is the uncertainty sampling calculation (corresponding to the CAD model construction module, the uncertainty sampling module, and the CFD batch calculation module) built based on a commercial design software CAESE, and the second part is the uncertainty analysis and multidisciplinary performance optimization based on the sampling results (corresponding to the DNN model training module, the uncertainty analysis module, and the turbine performance optimization module).

[0032] In the first part of the function, the CAD model construction module is used to parameterize the structure of the geometric model of the turbine blade. This system can collect the uncertainty sample set with one key. Specifically, in the CAD model construction module, import the model of a fully cooled turbine blade with a typical combination configuration of impingement and film cooling in the industry, and parameterize the structure of the geometric model of the turbine cooling blade, specifically including: the curvature of the blade inner cavity; the blade thickness distribution; the blade tip shape; the blade tail shape; the aperture, angle, distribution gap, and number of 11 exhaust film holes; the aperture, angle, distribution gap, and number of 20 rows of cooling holes; the thickness, distribution gap, and number of 4 groups of rib walls; the radius, distribution gap, and number of 4 groups of turbulators, etc. It basically covers all the cooling structures in the turbine cooling blade. Set the initial design value and appropriate value range for each parameter to ensure that the parameter values change within an acceptable range. As long as the geometric parameter values of the blade structure in the system are changed, the corresponding CAD model will also be adjusted accordingly. It improves the flexibility, efficiency, and accuracy of the design, and at the same time accelerates the design process, reduces complexity, and makes the design work more systematic and engineering-oriented. The parameterized model is shown in Figure 2 .

[0033] Secondly, by writing scripts to call the interfaces provided by the CAESES software, the interaction with other software is realized, and the integration of the CAD model construction module, the uncertainty sampling module, and the CFD batch calculation module is achieved based on the CAESES software. The uncertainty sampling module is used to select geometric parameters according to the research objectives and determine the uncertainty range of the geometric parameters according to the manufacturing and usage conditions, and sample within the uncertainty range of the geometric parameters; the CFD batch calculation module is used to batch process the sampling results of the uncertainty sampling module, specifically including mesh generation, preprocessing, solution, and postprocessing.

[0034] The CAESES software calls ANSYS ICEM for mesh generation, CFX-Pre for preprocessing, CFX-Solver for solution, CFX-Post for postprocessing, etc. The full process is as follows:

[0035] (1) Model creation and import: Create a CAD model according to the geometry of the actual flow problem and import it into the fluid calculation software, including the solid domain and the fluid domain.

[0036] (2) Mesh generation: Mesh the geometric model to discretize the solution domain. Commonly used mesh generation methods include structured meshes and unstructured meshes.

[0037] (3) Preprocessing: Define the initial and boundary conditions, including information such as the flow velocity, temperature, and pressure at the fluid inlet, as well as the boundary conditions at the fluid outlet or other surfaces; define the material properties.

[0038] (4) Solver solution: Select a suitable numerical method and solver to numerically solve the fluid equations, perform fluid simulation calculations, and monitor the convergence situation.

[0039] (5) Post-processing: Obtain the simulation calculation result data and perform statistics and visualization on it.

[0040] During the process of the uncertainty sampling module sampling within the uncertainty range of the geometric parameters, in order to reduce the number of cases required, the Latin Hypercube Sampling (LHS) method is used for uncertainty sampling calculation. This sampling method evenly divides the sample space into continuous regions, and selects a sample value from each region to construct the sampling set. Compared with the traditional random sampling method, LHS has the advantages of high sampling efficiency, strong coverage, and small sampling deviation.

[0041] Finally, based on the design optimization function of CAESES itself, before using the uncertainty sampling module for sampling, appropriate parameter selection and uncertainty range setting are required. Then, after determining the required output results and the number of data to be collected, one can perform sampling calculations with one key. The full process is as follows:

[0042] (1) Determination of uncertainty quantification target parameters: Determine the objectives of uncertainty research, select appropriate parameters or variables, and the uncertainty variation range.

[0043] (2) Establishment of parameter distribution function: Set a suitable distribution function for the selected parameters.

[0044] (3) Selection of sampling method to generate samples: The uncertainty sampling module uses the Latin Hypercube Sampling method to generate a certain number of samples to represent its uncertainty range.

[0045] (4) Simulation calculation and result collection: Apply the sample set of uncertainty parameters to the CFD batch calculation module and perform a large number of simulation calculations.

[0046] In the second part, a deep neural network (DNN) is selected as the surrogate model for uncertainty analysis, and a DNN model with one input layer, two hidden layers, and one output layer is constructed. Important blade geometric parameters are selected as inputs, and the physical quantities related to the blades of concern are selected as outputs; the arctangent function tanh is selected as the activation function for non-linear mapping; the root mean square error function RSME is selected as the loss function to measure the error between the model prediction output and the true value. The expression is as follows:

[0047]

[0048] Where n is the number of samples, y i is the true value of the sample, is the model's predicted value.

[0049] When training the DNN model with the sample set obtained by the uncertainty sampling module, the Grid Search method is used to optimize the combination of the number of nodes in the two hidden layers of the constructed DNN model. Other hyperparameters: learning rate: controls the step size or size of parameter update in each iteration, set to 0.01; weight decay: penalizes the weight parameter to prevent model overfitting, set to 0.1. On the other hand, even though the uncertainty sampling module greatly improves the efficiency of uncertainty sampling calculation, the CFD simulation calculation of each set of geometric parameters is extremely computationally intensive. In the face of the problem of data sparsity in the training of DNN models, the K-fold Cross Validation method is considered for model evaluation. The principle is to divide the data set into K equal subsets, one of which is the validation set each time, and the remaining K-1 subsets are the training sets. Repeat the cycle K times to traverse all subsets, and take the K average of the final evaluation results. This method can make better use of data and reduce the deviation of model evaluation results. This system uses the determination coefficient R 2 As an indicator for evaluating the degree of fit of the regression model, the calculation formula is as follows.

[0050]

[0051] where y i is the true value of the sample, is the model prediction value, and y is the true mean value of the sample.

[0052] The training steps of the DNN model are as follows:

[0053] (1) Search for node combinations in a large range to narrow down the appropriate node range. Select a larger node range and evaluate the degree of fit of the DNN model with different combinations (training set R 2 Mean) and generalization ability (validation set R 2 ) Keep the number of nodes within a reasonably small range.

[0054] (2) Determine the narrowed node search range and eliminate problematic data based on K-fold cross validation. Since CFD calculations inevitably result in large errors in calculation results due to issues such as model quality and mesh quality, such data will interfere with the actual physical input-output relationship during model training. Therefore, it is necessary to eliminate problematic data through a posteriori approach.

[0055] (3) Select the optimal node combination according to the training results. Comprehensively evaluate the R 2 mean value of different node combinations and select the optimal node combination based on the RSME value of the loss function.

[0056] After the DNN model is trained, use the uncertainty analysis module to perform sensitivity analysis and dimensionality reduction on the selected geometric parameters. Specifically, the SHAP (Shapley Additive Explanations) value analysis method is used for sensitivity analysis. The Shapley value is defined as the average marginal contribution rate of a certain eigenvalue in all possible combinations. For the i-th input feature vector x i to the output vector Y(x i ), the derivation of the Shapley value calculation formula is as follows:

[0057] Assume that S represents the subset of all feature vectors, N represents the set of all feature vectors, and x i represents the i-th input feature vector (specifically a vector composed of geometric parameters). The marginal contribution of adding x i to the subset S is:

[0058] δ i (S) = EY k (S ∪ (x i )) - EY k (S)

[0059] The prediction of the contribution value of removing the eigenvalue from S:

[0060] EY k (S) = E(Y(x k ))|S)

[0061] Calculate the consistent δ i (S) for all subsets. By weighted summing all possible combinations of eigenvalues, the final Shapley value formula can be derived:

[0062]

[0063] It is not difficult to see from the above formula derivation that by calculating the SHAP value and finding the average change in the prediction result after each feature is added, the influence degree of each feature on the prediction result can be understood, so as to better understand the decision-making process of the model and the importance of features.

[0064] The selected geometric parameters are analyzed for sensitivity using the uncertainty analysis module. The SHAP values of all features are calculated using the Kernel Explainer in the SHAP library of Python. The Kernel Explainer is used to explain any machine learning model and estimates the sensitivity of input features through an approximation method in the feature space. Since the SHAP value calculation method involves calculating for all possible feature subsets, when the number of features is large, all possible feature subsets are large, and direct calculation is very time-consuming. This system adopts an approximate calculation method, that is, estimates the Shapley value through random sampling, and then approximates the true Shapley value through the sample mean. The number of iterations, that is, the number of random samplings, is set to 100. Generally, SHAP mean plots and SHAP value summary plots are used for model interpretation. The former is the result of averaging the absolute values of the SHAP values for each feature value and can be regarded as a ranking plot of feature importance; the latter is a visual display of the SHAP values of all features. Each point represents a SHAP value, and the color represents the value size (increasing gradually from blue to red). The abscissa represents the SHAP value size, and the ordinate represents all features, that is, the features with a wider horizontal arrangement in the figure have greater influence. As Figure 3 , it is the SHAP value analysis diagram corresponding to a certain set of geometric parameters.

[0065] Considering that there may be a large number of selected geometric parameters and the existence of unimportant parameters, resulting in insufficient neural network accuracy due to the sparsity of the dataset, the sensitivity size relationship between two adjacent parameters in the SHAP mean plot cannot be determined. On the other hand, due to the complex geometric features of the turbine cooling blade, it is necessary to improve the calculation efficiency, so it is necessary to retain only the information of key features. This system adopts an inner loop method to reduce the dimension of the selected geometric parameters. The graphical illustration is shown in Figure 4 .

[0066] (1) Use the selected geometric parameters for training and optimize step by step to reduce the deviation between the R 2 value and the predicted average temperature value of the inner wall of the blade.

[0067] (2) Conduct SHAP value analysis on the selected parameters and rank their sensitivity to the inner wall temperature of the blade. And delete the geometric parameters with too low sensitivity.

[0068] (3) Retrain and optimize the neural network with the filtered geometric parameters and observe whether the quality of the neural network is optimized.

[0069] (4) If the quality of the neural network is optimized, it means that the deleted geometric parameters have too little influence on the output physical quantity, and the dimension reduction is achieved, and the loop continues; if the quality of the neural network deteriorates, it means that the deleted geometric parameters are relatively important, and the dimension reduction fails, and the loop ends and the dimension reduction process is completed.

[0070] For example, dimensionality reduction is performed based on the 29 groups of geometric parameters selected in the numerical example. The training set R of the K-fold cross-validation of the optimal node combination neural network during the dimensionality reduction cycle 2 and the validation set R 2 The changes in the mean value and the maximum value are shown in Figure 5 . From Figure 5 it can be seen that the first dimensionality reduction and the second dimensionality reduction both improve the quality of the neural network to a certain extent. When deleting parameters for the third time, the quality of the neural network drops significantly. It is concluded that when the 29 groups of geometric parameters of the turbine cooling blade selected in this numerical example are reduced to 8 groups, both the model accuracy and the parameter dimension meet the requirements. If the dimensionality reduction continues, important physical quantities are missing, and the nonlinearity and geometric integrity of the system cannot be reflected. Therefore, 8 groups of turbine cooling geometric parameters are selected as the final input according to the final effect.

[0071] Finally, based on the key geometric parameters after dimensionality reduction and the DNN model, deterministic optimization and uncertainty optimization are carried out to propose an optimization scheme for improving the turbine cooling performance. Considering performance and cost and other objectives comprehensively, and combining the results of uncertainty quantification, the optimal design of turbine cooling performance is realized.

[0072] This system performs the final multidisciplinary optimization work based on the genetic algorithm. Multidisciplinary Uncertainty Optimization (MUO) is an optimization method widely used in the fields of engineering and science, aiming to comprehensively consider the interactions of multiple disciplines, taking into account both design variables and uncertainties in the system, and optimizing the objective function. Multidisciplinary Design Optimization (MDO) is the core foundation of MUO, emphasizing collaborative design and optimization between different disciplines. In a multidisciplinary system, design variables and performance indicators usually have couplings between disciplines. In the research object of the present invention, changing the structural parameters may affect the fluid mechanics characteristics, which in turn will affect the heat transfer and performance of the system.

[0073] The Genetic Algorithm (GA) is an optimization method that simulates the biological evolution process. The core idea is to perform processes such as natural selection, genetic variation, and crossover to search for the optimal solution in the solution space of the problem. The genetic algorithm can be used to solve complex optimization problems and has strong applicability to nonlinear, high-dimensional, multi-peak, and difficult-to-solve optimization problems.

[0074] For example, considering the trade-off between the two objectives of the selected average blade surface temperature and the ratio of cooling gas to inlet flow rate, it is desired to find a solution that achieves as low a temperature as possible while minimizing the use of cooling gas. Two trained DNN models (one DNN model with the average temperature of the blade surface as the output and the other DNN model with the ratio of cooling gas to inlet flow rate as the output) are used to predict the average temperature of the blade surface and the ratio of cooling gas to inlet flow rate. By predicting the combination of values of each output physical quantity through the DNN model, the entire design space is effectively sampled. As Figure 6 shown, the Pareto optimal solutions are identified by searching for points in the sample space and simultaneously minimizing the two objectives (the average temperature of the blade surface and the ratio of cooling gas to inlet flow rate). These solutions form a set of Pareto front design points, as Figure 6 shown by the red points in. The genetic algorithm (GA) is adopted to iteratively approach the Pareto optimal solution by utilizing the crossover and mutation operations within the initial population to achieve a more accurate fitting of the Pareto front. Through the evolutionary process, the initial sample set gradually converges to the Pareto front, effectively approximating the Pareto curve, as Figure 7 shown. The algorithm parameters are set as follows: crossover probability CXPB = 0.8, mutation probability MUTPB = 0.1, number of generations NGEN = 200, and population size pop = 1000.

Claims

1. A turbine blade geometric structure uncertainty quantitative analysis system, characterized in that: It includes a CAD model building module, an uncertainty sampling module, a CFD batch calculation module, a DNN model training module, an uncertainty analysis module and a turbine performance optimization module; the CAD model building module is used to parameterize the structure of the geometric model of the turbine blade; the uncertainty sampling module is used to select geometric parameters according to the research objectives, and determine the uncertainty range of the geometric parameters according to the manufacturing and use conditions, and perform sampling within the uncertainty range of the geometric parameters; The CFD batch calculation module is used to perform batch processing on the sampling results of the uncertainty sampling module, specifically including grid division, preprocessing, solving and post-processing; the DNN model training module is used to train the DNN model according to the calculation results of the CFD batch calculation module, and the input of the DNN model is geometric parameters and the output is target physical quantities; the uncertainty analysis module is used to perform sensitivity analysis on the geometric parameters selected by the uncertainty sampling module according to the target physical quantities, and extract key geometric parameters; the turbine performance optimization module is used to perform deterministic optimization and uncertainty optimization on the key geometric parameters.

2. The turbine blade geometric structure uncertainty quantitative analysis system according to claim 1, characterized in that: The geometric model of the turbine blade is a model of a fully cooled turbine blade with a combined impingement and film cooling configuration.

3. The turbine blade geometric structure uncertainty quantitative analysis system according to claim 1, characterized in that: The CAD model building module is used to parameterize the structure of the geometric model of the turbine blade, wherein the structure of the geometric model of the turbine blade specifically includes: blade inner cavity curvature; blade thickness distribution; blade tip shape; blade tail shape; air film hole aperture, angle, distribution gap, number; impact hole aperture, angle, distribution gap, number; rib wall thickness, distribution gap, number; spoiler column radius, distribution gap, number.

4. The turbine blade geometric structure uncertainty quantitative analysis system according to claim 1, characterized in that: The deterministic optimization is specifically to select the best set of key geometric parameters according to the target physical quantity; the uncertain optimization is specifically to select a set of key geometric parameters according to the performance dispersion under the uncertainty brought about by manufacturing and use, so as to maximize the robustness of the turbine system.

5. The turbine blade geometric structure uncertainty quantitative analysis system according to claim 1, characterized in that: The meshing is specifically to mesh the geometric model of the turbine blade to discretize the solution domain; the preprocessing specifically includes defining initial and boundary conditions and material properties; the solution is specifically to numerically solve the fluid equations; and the post-processing is to statistically analyze and visualize the numerical solution results.

6. The turbine blade geometric structure uncertainty quantitative analysis system according to claim 1, characterized in that: The DNN model training module is used to train a DNN model according to the calculation results of the CFD batch calculation module. The specific method is: based on K-fold cross-validation, the calculation results of the CFD batch calculation module are randomly divided into K folds, and each fold is used as a validation set to perform multiple rounds of training on the DNN model; based on the Grid Search method, the hyperparameters of the hidden layer node combination of the DNN model are optimized; based on the key geometric parameters extracted by the uncertainty analysis module, the key geometric parameters are used as new DNN model inputs to continue training until the model accuracy reaches above the set threshold and the retained parameter dimensions meet the calculation requirements.

7. The turbine blade geometric structure uncertainty quantitative analysis system according to claim 1, characterized in that: The uncertainty analysis module is used to perform sensitivity analysis on the geometric parameters selected by the uncertainty sampling module based on the SHAP analysis method and extract key geometric parameters. The specific method is as follows: Calculate the SHAP value of the geometric parameter selected by the uncertainty sampling module. The larger the SHAP value of the geometric parameter, the greater its sensitivity. According to the SHAP value, the geometric parameter with greater sensitivity is selected as the key geometric parameter. The specific calculation formula of the SHAP value of the geometric parameter is: δ i (S)=EY k (S∪(x i ))-EY k (S) EY k (S)=E(Y(x k )|S) Among them, x i represents the i-th input feature vector, which is a vector consisting of geometric parameters; X is the vector that does not contain x i The set of all other input feature vectors, N represents the set of all feature vectors; Y(x k ) is x k The corresponding output vector.

Images