A method, system, equipment, and medium for optimizing the geometric parameters of a turbine blade.

Abstract

The present application relates to the technical field of gas turbine, and discloses a turbine blade geometric parameter optimization method, system, device and medium.The method comprises the following steps: modeling the turbine blade according to the geometric parameters, working condition parameters and material parameters of the turbine blade, obtaining a plurality of first input samples with a first fidelity and a plurality of second input samples with a second fidelity; obtaining the performance parameters of the turbine blade corresponding to each first input sample and each second input sample respectively as first response samples and second response samples; training a gas turbine turbine performance prediction network, and predicting the performance parameters of the target turbine blade according to the geometric parameters, working condition parameters and material parameters of the target turbine blade through the trained gas turbine turbine performance prediction network, and optimizing the geometric parameters of the target turbine blade with the maximum reliability and robustness of the target turbine blade indicated by the predicted performance parameters as the target.

Description

Technical Field

[0001] This invention relates to the field of gas turbine technology, and in particular to a method, system, device and medium for optimizing the geometric parameters of turbine blades. Background Technology

[0002] Gas turbines are core power equipment in modern industry, characterized by high-efficiency power generation, rapid response, and clean energy utilization. They are an important component of modern power systems. At the same time, gas turbines can provide powerful power for aircraft engines, ship propulsion, and oil and gas pipeline transportation. They are also a reliable power system for military equipment such as fighter jets and warships, and are of great significance to national energy security and defense security.

[0003] Gas turbine blades operate in harsh environments with high temperatures and high speeds for extended periods. Furthermore, due to grid fluctuations, changes in operating conditions, and the influence of actual environmental factors, the operating conditions of the blades are subject to various uncertainties. These uncertainties not only increase the risk of blade failure under extreme conditions but also lead to performance degradation under high-frequency conditions. Therefore, optimizing the reliability design of the blades is crucial for ensuring the safe and efficient operation of the unit.

[0004] However, in the robust optimization problem of turbine blade reliability, the input includes multiple uncertain parameters such as geometry and operation, and the output covers multiple response parameters such as aerodynamic performance, strength characteristics, and fatigue life. The method of obtaining a high-precision surrogate model by building a database requires a lot of time and computing resources to obtain sufficient sample data, resulting in high computational costs. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, device and medium for optimizing the geometric parameters of turbine blades, which can solve the problem that the method of obtaining a high-precision surrogate model requires a lot of time and computing resources to obtain sufficient sample data, resulting in high computing costs.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide a method for optimizing the geometric parameters of a turbine blade, comprising the following steps:

[0007] Based on the geometric parameters, operating parameters, and material parameters of the turbine blade, a model is created to obtain multiple first input samples with a first fidelity and multiple second input samples with a second fidelity; wherein, the first fidelity is less than the second fidelity.

[0008] Obtain the performance parameters of the turbine blade corresponding to each first input sample and each second input sample, respectively, and use them as the first response sample and the second response sample;

[0009] A gas turbine performance prediction network, comprising a first fidelity prediction network and a correction network, is trained using multiple first input samples, multiple second input samples, multiple first response samples, and multiple second response samples. During training, the first fidelity prediction network is trained using the first input samples and the first response samples to predict the first response sample corresponding to the first input sample. The correction network is then trained using the first response sample predicted by the first fidelity prediction network and the second input samples to predict the second response sample corresponding to the first input sample.

[0010] The trained gas turbine performance prediction network predicts the performance parameters of the target turbine blade based on its geometric parameters, operating parameters, and material parameters. The geometric parameters of the target turbine blade are then optimized with the goal of maximizing the reliability and robustness of the target turbine blade as indicated by the predicted performance parameters.

[0011] Optionally, the gas turbine performance prediction network is trained using an active learning sampling method. The active learning sampling method aims to minimize the estimation error of the failure probability of the target turbine blade indicated by the second response sample predicted by the gas turbine performance prediction network, and determines the location and fidelity of the new sampling points. The performance parameters of the turbine blade are used to indicate the failure probability of the turbine blade.

[0012] Optionally, the active learning sampling method determines the location and fidelity of the new sampling points using the following formula:

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] In the formula, x For the input vector, t This indicates the fidelity of the gas turbine performance prediction network. and These represent the first fidelity and the second fidelity, respectively. For adaptive learning function, This represents the probability that the sample point is accurately identified by the gas turbine performance prediction network. c ( t Let be the cost function, representing the ratio of the acquisition cost of the first input sample to the cost of the second input sample. It is a cross-correlation function. Let be the probability density function. Let it be a distance function. The mean of the predicted values, The variance of the predicted values, x new For the newly added sample points and t new This corresponds to the fidelity.

[0018] Optionally, the trained gas turbine performance prediction network predicts the performance parameters of the target turbine blade based on its geometric parameters, operating parameters, and material parameters. The geometric parameters of the target turbine blade are then optimized with the goal of maximizing the reliability and robustness indicated by the predicted performance parameters. This optimization includes:

[0019] The trained gas turbine performance prediction network predicts the performance parameters of the target turbine blade under different operating conditions based on the geometric parameters, material parameters, and parameters of different operating conditions. The performance parameters include the power, efficiency, stress, and strain of the target turbine blade.

[0020] Based on the stress and strain of the target turbine blade, the failure probability of the target turbine blade is obtained;

[0021] The geometric parameters of the target turbine blade are optimized with the goal of maximizing the mean power and efficiency and minimizing the variance under different operating conditions, and with the failure probability of the target turbine blade as a constraint.

[0022] Optionally, the geometric parameters include the geometric parameters of the blade root region and the rounded corner region at the bottom of the blade; the operating parameters include the total gas inlet temperature, total gas inlet pressure, gas outlet flow rate, total cooling air inlet temperature, total cooling air inlet pressure, and rotational speed of the gas turbine; and the material parameters include the density, elastic modulus, Poisson's ratio, thermal conductivity, and thermal conductivity of the turbine blade.

[0023] Embodiments of the present invention also provide a turbine blade geometry parameter optimization system, comprising:

[0024] The input sample acquisition module is used to model the turbine blade based on its geometric parameters, operating parameters, and material parameters, and obtain multiple first input samples with a first fidelity and multiple second input samples with a second fidelity; wherein the first fidelity is less than the second fidelity.

[0025] The response sample acquisition module is used to acquire the performance parameters of the turbine blade corresponding to each first input sample and each second input sample, respectively, as the first response sample and the second response sample.

[0026] The network training module is used to train a gas turbine performance prediction network, which includes a first fidelity prediction network and a correction network, using multiple first input samples, multiple second input samples, multiple first response samples, and multiple second response samples. During training, the first fidelity prediction network is trained using the first input samples and first response samples to predict the first response sample corresponding to the first input sample. The correction network is trained using the first response sample predicted by the first fidelity prediction network and the second input samples to predict the second response sample corresponding to the first input sample.

[0027] The parameter optimization module is used to predict the performance parameters of the target turbine blade based on the geometric parameters, operating parameters, and material parameters of the target turbine blade using a trained gas turbine performance prediction network. The module optimizes the geometric parameters of the target turbine blade with the goal of maximizing the reliability and robustness of the target turbine blade indicated by the predicted performance parameters.

[0028] Embodiments of the present invention also provide a computer device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the above-described turbine blade geometry parameter optimization method.

[0029] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for optimizing the geometric parameters of a turbine blade.

[0030] The turbine blade geometric parameter optimization method provided by this invention has at least the following beneficial effects:

[0031] To address the robust optimization problem of turbine blade reliability, two types of sample data with different fidelity are constructed: a first input sample (low-fidelity sample data) with a first fidelity and its corresponding first response sample, and a second input sample (high-fidelity sample data) with a second fidelity and its corresponding second response sample. When training the gas turbine performance prediction network (i.e., the surrogate model), the low-fidelity prediction network is trained using the low-fidelity sample data and its response data to predict the response data corresponding to the low-fidelity sample data. Then, the predicted values ​​of the low-fidelity prediction network and the high-fidelity sample data are used to correct the predicted values, thereby outputting high-fidelity response data. In other words, the actual prediction is performed by the low-fidelity surrogate model, and only the high-fidelity sample data is used for correction. This fully utilizes the complementary advantages of the high-fidelity model data and the low-fidelity model data, improving processing efficiency and reducing computational costs while ensuring the model's prediction accuracy.

[0032] Based on this model, the geometric parameters of the turbine blades can be optimized to achieve reliability design optimization of the turbine blades, so as to maintain high efficiency and stable performance and solve the problem of robust optimization of turbine blade reliability. Attached Figure Description

[0033] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0034] Figure 1 A flowchart illustrating a method for optimizing the geometric parameters of a turbine blade provided by the present invention;

[0035] Figure 2 This invention provides a schematic diagram of the structure of a gas turbine performance prediction network.

[0036] Figure 3 A schematic diagram of the training process of a gas turbine performance prediction network provided by the present invention;

[0037] Figure 4 This invention provides a schematic diagram of a dual-layer circulation process for robustly optimizing the reliability of gas turbine blades. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0039] Currently, the optimization of the geometric parameters of turbine blades faces multiple challenges: First, its inputs include uncertain parameters from multiple sources such as geometry and operation, while the outputs cover multiple response parameters such as aerodynamic performance, strength characteristics, and fatigue life. The method of obtaining a high-precision surrogate model by building a database requires a lot of time and computational resources to obtain sufficient sample data, resulting in high computational costs. Second, the robustness optimization problem involves multiple objective parameters, and the reliability constraints involve solving probability integrals, which makes robust reliability optimization a high-dimensional, multi-peak nonlinear programming problem. Traditional optimization algorithms are inefficient and prone to getting trapped in local optima. Finally, the physical field data of gas turbines have characteristics such as ultra-high dimensionality, strong nonlinearity, and spatiotemporal coupling. Traditional surrogate models often directly map input parameters to performance parameters, resulting in low prediction accuracy and a lack of physical interpretability.

[0040] Therefore, establishing a fast, accurate, and efficient robust optimization method for the reliability of gas turbine blades is of great significance for shortening the blade design cycle, reducing the performance fluctuation of the turbine under the influence of multiple uncertain factors, and ensuring the safe and stable operation of the gas turbine.

[0041] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0042] One embodiment of the present invention relates to a method for optimizing the geometric parameters of a turbine blade. The specific process of the method for optimizing the geometric parameters of a turbine blade in this embodiment is as follows: Figure 1 As shown, it includes:

[0043] Step 101: Based on the geometric parameters, operating parameters, and material parameters of the turbine blade, model the turbine blade to obtain multiple first input samples with a first fidelity and multiple second input samples with a second fidelity; wherein, the first fidelity is less than the second fidelity.

[0044] Step 102: Obtain the performance parameters of the turbine blade corresponding to each first input sample and each second input sample, respectively, as the first response sample and the second response sample.

[0045] Step 103: Train the gas turbine performance prediction network, which includes a first fidelity prediction network and a correction network, using multiple first input samples, multiple second input samples, multiple first response samples, and multiple second response samples. During training, the first fidelity prediction network is trained using the first input samples and the first response samples to predict the first response sample corresponding to the first input sample. The correction network is trained using the first response sample and the second input sample predicted by the first fidelity prediction network to predict the second response sample corresponding to the first input sample.

[0046] Step 104: Using the trained gas turbine performance prediction network, predict the performance parameters of the target turbine blade based on its geometric parameters, operating parameters, and material parameters. Optimize the geometric parameters of the target turbine blade with the goal of maximizing the reliability and robustness of the target turbine blade indicated by the predicted performance parameters.

[0047] The following is a detailed description of the implementation details of the turbine blade geometric parameter optimization method in this embodiment. The following content is only for the convenience of understanding and is not necessary for implementing this solution.

[0048] In step 101, a reasonable numerical calculation method is first adopted to determine the weak points of the turbine blade. Based on the actual situation of the blade, the design space of the turbine blade's geometric parameters and the environmental space of its operating parameters and material parameters are determined. First, gas-thermal-structure calculations are performed on the turbine blade to identify the blade root region and the rounded corner region at the bottom of the blade body as weak points. Then, parametric modeling is performed on these weak points, forming the geometric parameter design space. D Its expression is as follows:

[0049] ;

[0050] In the formula, g For parameters of the leaf root region; r This refers to the parameters for the rounded corner area at the bottom of the blade; the superscripts - and + indicate the lower and upper bounds of the parameters, respectively.

[0051] Based on the actual operating conditions of the blades, determine the operating parameters, material parameters, and environmental space. C Its expression is as follows:

[0052] ;

[0053] In the formula, c These are operating parameters, including total gas inlet temperature, total gas inlet pressure, gas outlet flow rate, total cooling air inlet temperature, total cooling air inlet pressure, and engine speed. mThese are the parameters of the blade material, including density, elastic modulus, Poisson's ratio, thermal conductivity, and thermal conductivity.

[0054] Then, Latin hypercube sampling is used to generate high-fidelity and low-fidelity sample sets. Traditional surrogate models built from single-fidelity data sources often face a trade-off between accuracy and efficiency during training. Specifically, high-fidelity (HF) models, due to their accurate modeling of physical problems, can produce high-confidence results, such as fine-grid models in FEM, direct numerical models in CFD, and full-scale experimental studies. However, such high-fidelity models are often accompanied by high computational / experimental costs, severely limiting the sample size. Conversely, low-fidelity (LF) models introduce simplification methods, significantly reducing computational / experimental costs while sacrificing some accuracy, such as using coarse grids, empirical formulas, and using characteristic simulations for experiments. Such low-fidelity (LF) models can quickly generate large amounts of data, but their inherent simplification errors can lead to systematic biases, especially significant when dealing with complex problems involving strong nonlinearity and multi-field coupling. In this embodiment, in the geometric parameter design space... D Operating parameters and material parameters, environmental space C Within this, a small initial sample set is constructed using the Latin hypercube sampling method. YB .

[0055] In step 102, numerical calculation methods are used to solve for the true responses of high-fidelity and low-fidelity samples respectively, and a training dataset is constructed. For the initial sample set established above... YB The true response of the sample is solved using a fine grid. R 1. Use a coarse grid to solve for the true response of the sample. R 2. The samples and response results are processed into the format required for network training. In this embodiment, the high-fidelity and low-fidelity samples differ only in mesh fineness. Both are responses calculated using geometric, operating, and material parameters to obtain the current sample's response, including turbine stage power and efficiency, turbine blade temperature field parameters, and stress and strain at weak points in the turbine blade. The training dataset format needs to be constructed according to the actual situation. In this embodiment, the low-fidelity prediction model LF is a fully connected deep neural network, and the high-fidelity prediction model HF is a graph convolutional neural network. The low-fidelity training dataset... T LF and high-fidelity training dataset T HF They are shown below:

[0056] ;

[0057] ;

[0058] In the formula, I The sample number matrixN This is a set of node location information and node attributes for each sample. A It is the set of adjacency matrices representing the connection relationships between nodes in each sample.

[0059] In step 103, a gas turbine performance prediction network based on a multi-fidelity hybrid network is constructed and trained. Multi-fidelity modeling leverages the complementary advantages of integrating high-fidelity (HF) and low-fidelity (LF) model data. Furthermore, traditional surrogate models are often limited by their prior assumptions, while deep neural networks still exhibit good prediction accuracy when facing high-dimensional, strongly nonlinear problems. Establishing a gas turbine performance prediction network based on a multi-fidelity hybrid network can provide a new solution to the accuracy-efficiency trade-off problem in the process of high-precision and rapid prediction of gas turbine performance. (See also...) Figure 2 The gas turbine performance prediction network GT-MFHN consists of two sub-networks connected in series, and is a low-fidelity prediction network. Used to predict values ​​for low-fidelity data, and to correct the network. This is used to fit the complex mapping relationship from low-fidelity data to high-fidelity data. In this embodiment, a low-fidelity prediction network... A fully connected deep neural network with multiple hidden layers is used. Its input is the design variables, and its output is the predicted low-fidelity turbine performance parameters. The specific mapping is as follows:

[0060] ;

[0061] In the formula, x The input vector includes geometric parameters, operating condition parameters, and material parameters. These are predicted values ​​for low-fidelity data, including turbine blade stress, strain, turbine power, and efficiency. For low-fidelity prediction networks The parameters to be learned in the text.

[0062] Correction network A graph convolutional neural network with multiple graph convolutional hidden layers is used, with the design variables and a low-fidelity prediction network as inputs. Predicted value The output is the predicted high-fidelity turbine performance parameters, specifically mapped as follows:

[0063] ;

[0064] In the formula, For high-fidelity data, the predicted value is... To correct the parameters to be learned in the network.

[0065] The gas turbine performance prediction network GT-MFHN was trained using supervised learning. The specific steps, based on a phased training strategy, are as follows:

[0066] S31, using a low-fidelity training dataset T LF Update the low-fidelity prediction network Learnable parameters Minimize the network's loss function Low-fidelity prediction network The training process is as follows:

[0067] ;

[0068] In the formula, For low-fidelity prediction networks The loss function is defined as the true value of low-fidelity data. Predicted values ​​compared to low-fidelity data The mean squared error is calculated, and a regularization loss is introduced, which is defined as follows:

[0069] ;

[0070] In the formula, M For low-fidelity training sample size, λ LF For low-fidelity prediction networks Regularization rate w LF For low-fidelity prediction networks Weighting coefficients. These are used in the loss function. By adding the product of the sum of squares of the network weight coefficients and the regularization rate, the model is forced to reduce parameter redundancy while minimizing prediction error, so as to balance the trade-off between model complexity and training error and prevent network overfitting.

[0071] S32: Preserving Low-Fidelity Prediction Networks Parameter freezing was used to run the offline proxy model, utilizing a high-fidelity training dataset. T HF Update and fix the network Learnable parameters Minimize the network's loss function Network correction The training process is as follows:

[0072] ;

[0073] ;

[0074] In the formula, To correct the network loss function, P To provide a high-fidelity training sample size, For the true value of high-fidelity data, For high-fidelity data, the predicted value is... λ HF To correct the network Regularization rate w HF To correct the network Weighting coefficients.

[0075] S33: Based on the active learning sampling method, an adaptive learning function AU suitable for multi-fidelity surrogate models is constructed. For reliability analysis problems, the core of active learning sampling lies in designing the optimal sampling strategy to enable the surrogate model to more accurately identify the boundary between the failure domain and the safety domain, i.e., the limit state surface, until the convergence criterion related to the accuracy of failure probability assessment is met. Its mathematical essence is to construct a Bayesian optimization problem with the goal of minimizing the failure probability estimation error. For the gas turbine performance prediction network GT-MFHN based on a multi-fidelity hybrid network, its active learning process should have a dual decision-making dimension: in addition to determining the location of new sample points, it is also necessary to determine the fidelity level required for updating samples, i.e., which fidelity model should be called during the active learning process, thereby maximizing the overall surrogate model accuracy improvement while minimizing the sample acquisition cost. To solve the above problems, this invention proposes an adaptive learning function AU suitable for multi-fidelity surrogate models, whose mathematical expression is as follows:

[0076] ;

[0077]

[0078] ;

[0079] In the formula, t Indicates model fidelity. c ( t Let be the cost function, representing the ratio of the acquisition cost of low-fidelity data to that of high-fidelity data. This represents the probability that a sample point is accurately identified as having a symbol by the current proxy model. It is a cross-correlation function. Let be the probability density function. Let it be a distance function. The mean of the predicted values, This represents the variance of the predicted values.

[0080] New sample points can be obtained based on the established adaptive learning function AU. x newand the corresponding fidelity t new Its expression is as follows:

[0081] ;

[0082] The convergence criterion for the adaptive learning function AU is equally important. An overly aggressive criterion can lead to insufficient prediction accuracy in the surrogate model, while an overly conservative criterion can result in unnecessary waste of computational resources. This invention uses a stability-based convergence criterion, where the predicted values ​​of high-fidelity data... If no significant changes occur during the iteration process, the model is considered to have converged and the addition of points is stopped. Its mathematical expression is as follows:

[0083]

[0084] in, The convergence threshold should be selected reasonably according to the actual situation. In this embodiment, the value is 0.01.

[0085] S34: The gas turbine performance prediction network GT-MFHN is trained using an adaptive learning function AU. (See also...) Figure 3 Based on the existing sample set, the low-fidelity prediction network obtained in S31 and S32 is used. and correction network The results are predicted, and the model's predictions are judged based on a stability convergence criterion. If the model fails to converge, an adaptive learning function AU is used to determine new sample points. x new and the corresponding fidelity t new The true response of the relevant fidelity model was calculated using numerical methods, added to the initial sample set, and the low-fidelity prediction network was retrained. and correction network The model prediction results are then re-evaluated using a stability convergence criterion; if the model converges, the current low-fidelity prediction network is saved. and correction network The learnable parameter results were used to complete the training of the gas turbine performance prediction network GT-MFHN.

[0086] In step 104, based on the trained gas turbine performance prediction network GT-MFHN, a two-layer recurrent framework is used to perform reliability robustness optimization of the gas turbine blades. (See also...) Figure 4This invention proposes a two-layer cyclic framework. The inner layer uses Monte Carlo (MC) sampling to obtain the mean and variance of turbine performance parameters, as well as the failure probability of turbine blades. The outer layer, based on the data obtained from the inner layer, optimizes the turbine geometry parameters by maximizing the mean and minimizing the variance of the turbine performance parameters, while setting the failure probability of the turbine blades as a constraint. The specific operation steps are as follows:

[0087] S41: Construct the inner Monte Carlo loop. This is based on the current input turbine blade root region parameters. g Parameters of the rounded corner area at the bottom of the leaf r Monte Carlo sampling (MC) was used to measure the environmental space of operating parameters and material parameters. C Internally, regarding operating parameters c Blade material parameters m Sampling is performed, and based on the high-precision gas turbine performance prediction network GT-MFHN trained in S4, rapid prediction is made for each operating point to obtain the power, efficiency, stress, strain and other parameters of the turbine blades at each operating point.

[0088] S42: Calculate the environmental space of the operating parameters and material parameters under the current geometric parameters. C The uncertainty response of turbine performance parameters is quantified using mean and variance. Simultaneously, based on relevant life prediction models, the lifespan of the turbine blades at each operating point is calculated, along with the failure probability of the turbine blades. The formulas for calculating the mean and variance of the performance parameters are as follows:

[0089] ;

[0090] ;

[0091] In the formula, For performance parameters l The mean of the uncertainty response, For performance indicators l Uncertainty response standard deviation, performance parameters l Including turbine power W and efficiency wait, The number of samples for Monte Carlo sampling (MC).

[0092] The fatigue life of gas turbine blades can be characterized as an implicit function of stress or strain. N f ( z If ), then its limit state function for fatigue reliability assessment g ( z This can be represented as:

[0093] ;

[0094] In the formula, The defined safe life represents the design life under a certain safety margin. Its specific value needs to be determined by a combination of factors such as design requirements, risk assessment, and historical data.

[0095] Failure probability P f It can be represented as:

[0096] ;

[0097] S43: Construct an outer-layer robust optimization loop. Select an appropriate optimization method, and based on the data obtained from the inner layer, use maximizing the mean and minimizing the variance of turbine performance parameters as the optimization objectives. Simultaneously, set the constraint as the failure probability of the turbine blades, and perform iterative optimization of the turbine blade geometry parameters. Turbine power. W and turbine efficiency The mathematical description of robust multi-objective optimization can be expressed as follows:

[0098] , ;

[0099] In the formula, The defined failure probability threshold.

[0100] A heuristic non-dominated sorting genetic algorithm is used for reliability robustness optimization. It divides individuals in the population into different non-dominated levels and calculates the crowding distance for individuals in each non-dominated level. d i The algorithm selects individuals to form a new parent population based on the current non-dominated level and crowding distance, iterating until a termination condition is met, ultimately obtaining the Pareto optimal solution set. It possesses powerful global search capabilities, effectively exploring the entire design space, and is naturally suitable for parallel computation under multiple operating conditions, exhibiting significant advantages in robust optimization problems related to the reliability of gas turbine blades. Among these advantages, the crowding distance... d i The expression is as follows:

[0101] ;

[0102] In the formula, and For individuals i In the objective function f k The function values ​​of adjacent individuals, and For each individual in the non-dominated level, the objective function is... fk The maximum and minimum values ​​on.

[0103] The relevant parameters in the optimization algorithm should be selected based on optimization efficiency and actual computing resources. In this embodiment, considering optimization efficiency and computational cost, the population size is set to 200, the maximum number of iterations is 100, the crossover probability is 0.9, the mutation probability is 0.1, and both the crossover distribution index and the mutation distribution index are 20.

[0104] S44: A dual-layer loop framework is used to optimize the performance of the gas turbine. When the optimal solution set has not changed or the maximum number of iterations has been reached, the optimization stops and the final optimal solution set is output.

[0105] The turbine blade geometry parameter optimization method of this embodiment has at least the following beneficial effects:

[0106] 1. By constructing an adaptive learning function, the relevant parameters and fidelity of newly added sample points are actively obtained, and the surrogate model is automatically updated until the accuracy requirements are met. The training process is highly versatile and requires no manual intervention, achieving efficient utilization of training samples and greatly reducing the time and computational costs required to construct the dataset.

[0107] 2. The multi-fidelity hybrid network includes a low-fidelity prediction network and a correction network. It fully utilizes the complementary advantages of high-fidelity model data and low-fidelity model data to achieve multi-source data fusion. It can provide a new solution to the accuracy-efficiency trade-off problem in the process of high-precision and rapid prediction of gas turbine performance. At the same time, the correction network considers the physical field information of the model, which is more accurate and more interpretable than traditional proxy models.

[0108] 3. A two-layer loop architecture is adopted for reliability robust optimization. The inner layer uses a trained multi-fidelity hybrid network to quickly obtain the response under different operating conditions and calculate the reliability and robustness parameters of the turbine blade. The outer layer performs reliability robust optimization of the blade in the entire design space based on the response of the inner layer. This can obtain a design scheme that combines aerodynamic robustness and strength reliability, ensuring the high efficiency and reliability of turbine operation.

[0109] 4. By employing an optimization method to conduct multi-objective optimization for multiple optimization variables in the reliability optimization problem, while maintaining the orthogonality of the objective functions, the Pareto optimal solution set can be obtained completely, providing decision-makers with a multi-dimensional optimization choice space. This method is characterized by fast computation speed, low complexity, and good convergence.

[0110] The steps of the various methods described above are only for clarity. In practice, they can be combined into one step or some steps can be split into multiple steps. As long as they include the same logical relationship, they are all within the protection scope of this invention. Adding insignificant modifications or introducing insignificant designs to the algorithm or process, without changing the core design of the algorithm and process, are also within the protection scope of this invention.

[0111] Another embodiment of the present invention relates to a turbine blade geometry parameter optimization system. The implementation details of this turbine blade geometry parameter optimization system are described below. The following content is for ease of understanding and is not essential for implementing this solution. The turbine blade geometry parameter optimization system of this embodiment includes:

[0112] The input sample acquisition module is used to model the turbine blade based on its geometric parameters, operating parameters, and material parameters, and obtain multiple first input samples with a first fidelity and multiple second input samples with a second fidelity; wherein the first fidelity is less than the second fidelity.

[0113] The response sample acquisition module is used to acquire the performance parameters of the turbine blade corresponding to each first input sample and each second input sample, respectively, as the first response sample and the second response sample.

[0114] The network training module is used to train a gas turbine performance prediction network, which includes a first fidelity prediction network and a correction network, using multiple first input samples, multiple second input samples, multiple first response samples, and multiple second response samples. During training, the first fidelity prediction network is trained using the first input samples and first response samples to predict the first response sample corresponding to the first input sample. The correction network is trained using the first response sample predicted by the first fidelity prediction network and the second input samples to predict the second response sample corresponding to the first input sample.

[0115] The parameter optimization module is used to predict the performance parameters of the target turbine blade based on the geometric parameters, operating parameters, and material parameters of the target turbine blade using a trained gas turbine performance prediction network. The module optimizes the geometric parameters of the target turbine blade with the goal of maximizing the reliability and robustness of the target turbine blade indicated by the predicted performance parameters.

[0116] It is not difficult to see that this embodiment is a system embodiment corresponding to the above method embodiments, and this embodiment can be implemented in conjunction with the above method embodiments. The relevant technical details and technical effects mentioned in the above embodiments are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above embodiments.

[0117] It is worth mentioning that all modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. Furthermore, to highlight the innovative aspects of this invention, this embodiment does not introduce units that are not closely related to solving the technical problem proposed by this invention; however, this does not mean that other units are absent from this embodiment.

[0118] Another embodiment of the present invention relates to a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the turbine blade geometry parameter optimization method of the above embodiments.

[0119] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0120] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0121] Another embodiment of the present invention relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the method embodiments described above.

[0122] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing the present invention, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A method of optimizing the geometry of a turbine blade, characterized in that, The method comprises: According to the geometric parameters, working condition parameters and material parameters of the turbine blade, a turbine blade is modeled to obtain a plurality of first input samples with a first fidelity and a plurality of second input samples with a second fidelity; wherein the first fidelity is less than the second fidelity; Obtain the performance parameters of the turbine blade corresponding to each first input sample and each second input sample as first response samples and second response samples, respectively; A gas turbine turbine performance prediction network comprising a first fidelity prediction network and a correction network is trained using a plurality of first input samples, a plurality of second input samples, a plurality of first response samples and a plurality of second response samples; wherein the first input sample and the first response sample are used to train the first fidelity prediction network to predict the first response sample corresponding to the first input sample through the first fidelity prediction network, and the first response sample predicted by the first fidelity prediction network and the second input sample are used to train the correction network to predict the second response sample corresponding to the first input sample through the correction network; Through the trained gas turbine turbine performance prediction network, the performance parameters of the target turbine blade are predicted according to the geometric parameters, working condition parameters and material parameters of the target turbine blade, and the reliability and robustness of the target turbine blade indicated by the predicted performance parameters are maximized as the target to optimize the geometric parameters of the target turbine blade; The gas turbine turbine performance prediction network is trained using an active learning sampling method, which determines the position of the new sampling point and the fidelity of the new sampling point with the minimum error of the failure probability estimation of the target turbine blade indicated by the second response sample predicted by the gas turbine turbine performance prediction network as the target; wherein the performance parameters of the turbine blade are used to indicate the failure probability of the turbine blade; The active learning sampling method determines the position of the new sampling point and the fidelity of the new sampling point by the following formula: ； ； ； ； In the formula, x For the input vector, t This indicates the fidelity of the gas turbine performance prediction network. and These represent the first fidelity and the second fidelity, respectively. For adaptive learning function, This represents the probability that the sample point is accurately identified by the gas turbine performance prediction network. c ( t Let be the cost function, representing the ratio of the acquisition cost of the first input sample to the cost of the second input sample. It is a cross-correlation function. Let be the probability density function. Let be the distance function. The mean of the predicted values, The variance of the predicted values, x new For the newly added sample points and t new This corresponds to the fidelity.

2. The method of optimizing geometrical parameters of a turbine blade according to claim 1, characterized in that, The trained gas turbine turbine performance prediction network is used to predict the performance parameters of the target turbine blade according to the geometric parameters, material parameters and different working condition parameters of the target turbine blade, and the reliability and robustness of the target turbine blade indicated by the predicted performance parameters are maximized as the target to optimize the geometric parameters of the target turbine blade, comprising: The trained gas turbine turbine performance prediction network is used to predict the performance parameters of the target turbine blade under different working conditions according to the geometric parameters, material parameters and different working condition parameters of the target turbine blade; wherein the performance parameters include the power, efficiency, stress and strain of the target turbine blade; According to the stress and strain of the target turbine blade, the failure probability of the target turbine blade is obtained; The mean value of the power and efficiency of the target turbine blade under different working conditions is maximized, and the variance is minimized as the target, and the failure probability of the target turbine blade is constrained to optimize the geometric parameters of the target turbine blade.

3. The method of optimizing geometrical parameters of a turbine blade according to claim 1 or 2, characterized in that, The geometry parameters include geometry parameters of a blade root region and a blade body bottom fillet region of the turbine blade, the working condition parameters include a total gas inlet temperature, a total gas inlet pressure, a gas outlet flow rate, a total cooling air inlet temperature, a total cooling air inlet pressure, and a rotating speed of the gas turbine, and the material parameters include a density, an elastic modulus, a Poisson's ratio, a thermal conductivity coefficient, and a thermal conductivity of the turbine blade.

4. A system for optimizing the geometry of a turbine blade, characterized in that The system comprises: An input sample acquisition module configured to model the turbine blade according to the geometry parameters, the working condition parameters, and the material parameters of the turbine blade, and to obtain a plurality of first input samples with a first fidelity and a plurality of second input samples with a second fidelity; wherein the first fidelity is less than the second fidelity; A response sample acquisition module configured to acquire performance parameters of the turbine blade corresponding to each of the first input samples and each of the second input samples as first response samples and second response samples; A network training module configured to train a gas turbine turbine performance prediction network comprising a first fidelity prediction network and a correction network using the plurality of first input samples, the plurality of second input samples, the plurality of first response samples, and the plurality of second response samples; wherein the first fidelity prediction network is trained using the first input samples and the first response samples to predict the first response samples corresponding to the first input samples through the first fidelity prediction network, and the correction network is trained using the first response samples predicted by the first fidelity prediction network and the second input samples to predict the second response samples corresponding to the first input samples through the correction network; A parameter optimization module configured to predict performance parameters of a target turbine blade according to the geometry parameters, the working condition parameters, and the material parameters of the target turbine blade through the trained gas turbine turbine performance prediction network, and to optimize the geometry parameters of the target turbine blade with a target of maximizing the reliability and robustness of the target turbine blade indicated by the predicted performance parameters; The gas turbine turbine performance prediction network is trained using an active learning sampling method, which aims to minimize the error of failure probability estimation of the target turbine blade indicated by the second response samples predicted by the gas turbine turbine performance prediction network to determine the position of the new sampling point and the fidelity of the new sampling point; wherein the performance parameters of the turbine blade are used to indicate the failure probability of the turbine blade. The active learning sampling method determines the position of the new sampling point and the fidelity of the new sampling point through the following formula: ； ； ； ； wherein, x is an input vector, t denotes the fidelity of the gas turbine performance prediction network, and denote the first and second fidelity, respectively, is an adaptive learning function, denotes the probability that a sample point is correctly judged by the gas turbine performance prediction network, c is a cost function, t denotes the cost ratio of the first and second input samples, is a cross-correlation function, is a probability density function, is a distance function, is the mean of the predicted values, is the variance of the predicted values, x new is a new sample point and t new is the corresponding fidelity.

5. A computer device, comprising: Comprises: At least one processor; and a memory in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the turbine blade geometry parameter optimization method of any one of claims 1 to 3.

6. A computer readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to implement the turbine blade geometry parameter optimization method of any one of claims 1 to 3. The computer program is executed by the processor to implement the turbine blade geometry parameter optimization method of any one of claims 1 to 3.

Images