A method and system for uncertainty analysis of turbine tip gas thermal performance

Abstract

This invention discloses a method and system for uncertainty analysis of turbine blade tip aero-thermal performance. The method first identifies key structural parameters of the turbine blade tip as uncertainty variables and constructs a multi-precision sample set. Then, a multi-fidelity deep neural network model is constructed using cascaded prediction and correction sub-networks, establishing a mapping correction relationship from low to high precision through phased training. Next, the trained blade tip aero-thermal performance prediction network model is used to analyze the aero-thermal performance distribution, and the influence of each uncertainty variable on performance fluctuations is quantified using the Sobol global sensitivity analysis method. Finally, significant parameters are sorted according to their influence and their optimal value combinations are determined, forming a design scheme that makes the turbine blade tip aero-thermal performance approach the global optimum. This invention effectively overcomes the dimensional limitations of high-dimensional uncertainty analysis, significantly improves computational efficiency and prediction accuracy, and provides reliable technical support for robust turbine blade tip design.

Description

Technical Field

[0001] This invention relates to the field of turbine aerodynamic performance analysis technology, specifically to an uncertainty analysis method for the thermal performance of turbine blade tip gas under high load. Background Technology

[0002] Gas turbines and other power plants such as aero engines have compact overall structures, with the turbine tip, a core component, having particularly small geometric dimensions. These dimensions present significant manufacturing challenges, as even minute deviations can lead to substantial changes in the turbine tip's shape. Furthermore, transonic turbine tip blades directly withstand the impact of high-temperature, high-pressure combustion gases during high-speed rotation. Combined with the rapid changes in operating parameters during start-up and shutdown, these factors cause fluctuations in the turbine tip's operating state, deviating it from ideal design conditions. These uncertainties significantly affect the aerothermal performance of the turbine tip. Therefore, during turbine tip design, it is necessary to comprehensively consider the uncertainties in geometry and operating conditions, accurately predicting their impact on the turbine tip's aerothermal performance, thereby providing a reliable guarantee for the safe and stable operation of the gas turbine.

[0003] With the continuous development of uncertainty quantification methods, existing multi-fidelity methods (such as polynomial chaotic expansion methods) realize the mapping relationship between uncertain inputs and statistical characteristics by constructing orthogonal polynomial bases. However, their core limitation lies in their strong dependence on the input distribution and their susceptibility to the curse of dimensionality when facing high-dimensional uncertainty analysis. The curse of dimensionality has become a key bottleneck restricting the widespread application of traditional methods in engineering practice. Taking the analysis of turbine blade tip gas thermal performance as an example, when the dimension of uncertain variables increases significantly, the number of high-precision samples required by the polynomial chaotic expansion method increases exponentially, leading to a sharp increase in computational resource demands. Especially in analyses involving high-dimensional parameter spaces (such as 15 dimensions or more), maintaining prediction accuracy requires simulation resources far exceeding the actual engineering capacity, thus creating an irreconcilable contradiction between computational efficiency and model accuracy. Therefore, how to overcome resource constraints in high-dimensional uncertainty quantification and construct analysis methods that balance efficiency and reliability has become a key problem that urgently needs to be solved in the field of engineering optimization. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a method and system for uncertainty analysis of turbine blade tip aero-thermal performance. This method can effectively overcome the dimensional limitations of uncertainty analysis, significantly improve computational efficiency, and increase the uncertainty quantification accuracy of high-load turbine blade tips to over 98%, providing strong support for the efficient and stable design of aero-engines.

[0005] This invention is achieved through the following technical solution:

[0006] In a first aspect, this application provides a method for uncertainty analysis of the thermal performance of turbine blade tip gas, comprising the following steps:

[0007] Step 1: Determine the key structural parameters of the turbine blade tip and use them as uncertainty variables. Based on the uncertainty variables, generate a sample set containing sample points with different precision, determine the aero-thermal performance of each sample point, and then construct a first training dataset and a second training dataset with different precision.

[0008] Step 2: Construct a blade tip gas thermal performance prediction network model using cascaded prediction sub-networks and correction sub-networks;

[0009] The prediction subnetwork is trained using the first training dataset to learn the global trend of air-thermal performance. Then, the correction subnetwork is trained together with the second training dataset and the output of the prediction subnetwork to establish a mapping correction relationship from low precision to high precision until the loss function is minimized, thus obtaining the trained blade tip air-thermal performance prediction network model.

[0010] Step 3: Generate analysis sample points based on the uncertain variables, use the trained blade tip air-thermal performance prediction network model to predict air-thermal performance, obtain the air-thermal performance probability distribution through statistical analysis of the prediction results, and screen out analysis sample points whose air-thermal performance probability exceeds the preset threshold.

[0011] Step 4: Based on the Sobol global sensitivity analysis method, determine the degree of influence of each uncertainty variable on the fluctuation of gas-thermal performance in the analysis sample points;

[0012] Step 5: Sort all uncertain variables by their degree of influence, select significant parameters based on preset thresholds, determine the optimal values ​​of each significant parameter, and combine them to form a parameter combination that makes the turbine tip gas thermal performance approach the global optimum.

[0013] Preferably, in step 1, determining the gas-thermal properties of each sample point specifically involves:

[0014] A three-dimensional model of the turbine blade tip corresponding to each sample point is established. The gas-thermal performance of the sample points is obtained by simulating the three-dimensional model of the turbine blade tip through fluid dynamics. The gas-thermal performance includes the total pressure loss coefficient and the area-average heat transfer coefficient.

[0015] Preferably, before step 2, the following steps are also included:

[0016] A Gaussian process surrogate model is constructed based on the sample points in the second training dataset. A stochastic process mapping relationship between the sample points and the gas-thermal performance is established. The expected improvement function is used as the acquisition function. New high-precision sample points are generated iteratively through a Bayesian optimization strategy. The gas-thermal performance of the new high-precision sample points is calculated by simulation and the Gaussian process surrogate model is updated until the Gaussian process surrogate model converges. The new high-precision sample points are then merged into the second training dataset.

[0017] Preferably, establishing the stochastic process mapping relationship between sample points and gas thermal properties includes:

[0018] A stochastic process mapping relationship between input sample points and gas-thermal properties is established based on radial basis functions.

[0019] Preferably, the step of generating analysis sample points based on the uncertainty variables includes:

[0020] Monte Carlo sampling is performed in the multidimensional parameter space composed of uncertain variables according to the probability distribution of each uncertain variable.

[0021] For key structural parameters that follow a Gaussian distribution, Box-Muller transformation is used to generate normal random numbers, while for key structural parameters that follow a uniform distribution, inverse transformation is used to sample and generate analysis sample points.

[0022] Preferably, the method based on Sobol global sensitivity analysis, determining the degree of influence of each uncertainty variable on the fluctuation of gas-thermal performance in the analysis sample points, includes:

[0023] One key structural parameter among the uncertain variables is treated as a single design parameter, while the other key structural parameters are integrated into a constant within their parameter range.

[0024] Based on the trained blade tip air-thermal performance prediction network model, the weighted average value of the air-thermal performance corresponding to each value of a single design parameter is calculated. Using the air-thermal performance as the objective function, a single relationship model between a single design parameter and air-thermal performance is constructed.

[0025] The ratio of the variance corresponding to a single design parameter to the total variance is calculated based on the single-relationship model to obtain the degree of influence of the single design parameter on the gas-thermal performance.

[0026] Preferably, it also includes an analysis of the interactive effects of multiple design parameters on gas-thermal performance:

[0027] Multiple key structural parameters among the uncertain variables are treated as multiple design parameters, and other key structural parameters are integrated into constants within their parameter ranges.

[0028] Based on the trained blade tip air-thermal performance prediction network model, the weighted average of the air-thermal performance corresponding to each value of multiple design parameters is calculated. Using air-thermal performance as the objective function, an interaction model of multiple design parameters on air-thermal performance is constructed.

[0029] The ratio of the coupling variance to the total variance of multiple design parameters is calculated based on the interaction model, thus obtaining the degree of influence of the interaction of multiple design parameters on gas-thermal performance.

[0030] Preferably, determining the optimal values ​​of each significant parameter includes:

[0031] Based on the Sobol global sensitivity analysis method, the ratio of the variance of each design parameter to the total variance is determined. The influence is ranked according to the ratio, significant parameters are screened according to the ratio threshold, the optimal value of the significant parameters is determined according to the single relation model, and the optimal parameter combination is constructed according to the optimal value.

[0032] Secondly, this application provides an uncertainty analysis system for the thermal performance of turbine blade tip gas, comprising:

[0033] The acquisition module is used to determine the key structural parameters of the turbine blade tip and use them as uncertainty variables. Based on the uncertainty variables, a sample set containing sample points with different precision is generated, the aero-thermal performance of each sample point is determined, and then a first training dataset and a second training dataset with different precision are constructed.

[0034] The training module is used to construct a prediction network model for the thermal performance of the blade tip using cascaded prediction and correction subnetworks.

[0035] The prediction subnetwork is trained using the first training dataset to learn the global trend of air-thermal performance. Then, the correction subnetwork is trained together with the second training dataset and the output of the prediction subnetwork to establish a mapping correction relationship from low precision to high precision until the loss function is minimized, thus obtaining the trained blade tip air-thermal performance prediction network model.

[0036] The statistical module is used to generate analysis sample points based on the uncertain variables, use the trained blade tip air-thermal performance prediction network model to predict air-thermal performance, obtain the air-thermal performance probability distribution through statistical analysis of the prediction results, and screen out analysis sample points whose air-thermal performance probability exceeds a preset threshold.

[0037] The analysis module is used to determine the degree of influence of each uncertainty variable on the fluctuation of gas-thermal performance in the analysis sample points based on the Sobol global sensitivity analysis method.

[0038] The optimization module is used to sort all uncertain variables by their degree of influence, filter out significant parameters based on preset thresholds, determine the optimal values ​​of each significant parameter, and combine them to form a parameter combination that makes the turbine tip air thermal performance approach the global optimum.

[0039] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the uncertainty analysis method for the thermal performance of turbine blade tip gas.

[0040] Compared with the prior art, the present invention has the following beneficial technical effects:

[0041] This application provides an uncertainty analysis method for turbine blade tip aero-thermal performance. First, training datasets of varying precision are constructed, employing a cascaded prediction sub-network-correction sub-network architecture. The prediction sub-network learns the global variation patterns of performance from low-precision data, while the correction sub-network uses high-precision data for precise correction. This collaborative mechanism significantly reduces reliance on simulation data while maintaining prediction accuracy. Based on the trained blade tip aero-thermal performance prediction network model, Monte Carlo sampling and Sobol global sensitivity analysis methods are further combined. This not only achieves efficient acquisition of the probability distribution of aero-thermal performance but also quantitatively analyzes the independent influence and interaction effects of various uncertainty parameters, thereby accurately identifying the significant parameters contributing most to performance fluctuations. Finally, the sensitivity analysis results guide parameter optimization design, forming a parameter combination that brings the turbine blade tip aero-thermal performance close to the global optimum. This provides a novel solution for robust turbine blade tip design and has significant engineering application value.

[0042] This application also proposes an uncertainty analysis system for turbine blade tip gas thermal performance, an electronic device, and a computer storage medium, which possess all the advantages of the aforementioned uncertainty analysis method for turbine blade tip gas thermal performance. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a three-dimensional model of the turbine blade tip of the present invention;

[0045] Figure 2 This is a cross-sectional view of the geometric structure of the turbine blade tip of the present invention;

[0046] Figure 3 This is a flowchart of the uncertainty analysis method for the turbine blade tip gas thermal performance of the present invention.

[0047] In the diagram, 1 is the cooling chamber; 2 is the cooling channel; 3 is the groove; 4 is the air film hole; 5 is the bottom surface of the groove; 6 is the side wall of the groove; and 7 is the upper wall. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0049] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this invention.

[0050] Turbine blade tip refers to the top of the turbine blade. Figure 1 This is a 3D model of the turbine blade tip. The turbine blade has a hollow structure, which serves as the cooling chamber 1. The lower end of the cooling chamber 1 is connected to the cooling channel 2. A groove 3 is provided on the top of the turbine blade, and multiple film cooling holes 4 are spaced apart in the groove 3. The film cooling holes 4 are connected to the cooling chamber 1. The depth of the groove 3 is D, the width of the turbine blade tip shoulder is B, and the diameter of the film cooling holes is d. Figure 2 This is a cross-sectional view of the turbine blade tip geometry. In this view, a blade tip clearance is formed between the turbine blade tip and the upper end wall 7 of the gas turbine, and the height of the blade tip clearance is H.

[0051] An uncertainty analysis method for the thermal performance of turbine blade tip gas includes the following steps:

[0052] Step 1: Treat the key structural parameters of the turbine blade tip as uncertain variables, and determine the probability distribution form and value range of each uncertain variable;

[0053] This key structural parameter has a significant impact on the thermal performance of the turbine blade tip.

[0054] Step 2: In the multidimensional parameter space composed of uncertain variables, the Latin hypercube sampling method is used to generate a sample set containing sample points with different precision.

[0055] Sample points of different precision include high-precision sample points and low-precision sample points.

[0056] For each sample point in the sample set, a corresponding three-dimensional model of the turbine blade tip is established and fluid dynamics simulation is performed to obtain the aero-thermal performance of the turbine blade tip. The aero-thermal performance includes the total pressure loss coefficient characterizing aerodynamic performance and the area-average heat transfer coefficient characterizing heat transfer performance.

[0057] Based on the sample set and its corresponding gas and thermal properties, a first training dataset and a second training dataset are constructed.

[0058] The sample points in the first training dataset are low-precision sample points, while the sample points in the second training dataset are high-precision sample points.

[0059] Step 3: Construct a Gaussian process surrogate model based on high-precision sample points. Use the expected improvement function as the acquisition function. Iteratively generate new high-precision sample points through a Bayesian optimization strategy. Then, use simulation to calculate the gas-thermal performance of the new high-precision sample points to update the Gaussian process surrogate model until the Gaussian process surrogate model converges. Finally, merge the new high-precision sample points generated by the Gaussian process surrogate model into the second training dataset.

[0060] Step 4: Construct a blade tip air thermal performance prediction network model using a multi-fidelity deep neural network. The blade tip air thermal performance prediction network model includes cascaded prediction subnetworks and correction subnetworks.

[0061] The prediction sub-network is trained using the first training dataset to learn the global trend of air-thermal performance; the correction sub-network is trained together with the output of the prediction sub-network using the second training dataset to establish a mapping correction relationship from low-precision prediction to high-precision results until the loss function is minimized, thus obtaining the trained blade tip air-thermal performance prediction network model.

[0062] Step 5: Perform Monte Carlo sampling in the multidimensional parameter space to generate analysis sample points. Use the trained blade tip gas-thermal performance prediction network model to predict the gas-thermal performance of each analysis sample point. Perform statistical analysis on all prediction results to obtain the probability distribution of gas-thermal performance and screen out the analysis sample points whose gas-thermal performance probability exceeds the preset threshold.

[0063] Step 6: Based on the Sobol global sensitivity analysis method, determine the degree of influence of each uncertainty variable on the fluctuation of gas-thermal performance in the analysis sample points.

[0064] Step 7: Sort all uncertain variables by their degree of influence, select significant parameters based on preset thresholds, determine the optimal values ​​of the significant parameters, and form a parameter combination that makes the turbine tip gas thermal performance approach the global optimum based on all the optimal values.

[0065] Example 1

[0066] An uncertainty analysis method for the thermal performance of turbine blade tip gas includes the following steps:

[0067] Step 1: Treat the key structural parameters of the turbine blade tip as uncertain variables, and determine the probability distribution form and value range of the uncertain variables.

[0068] S1.1. Based on the geometric characteristics of the turbine blade tip, key structural parameters are selected as uncertainty variables.

[0069] Figure 2 This is a cross-sectional view of the geometric structure of the turbine blade tip. In this embodiment, the number of uncertain variables selected is 5, namely 5 key structural parameters, including the height H of the blade tip clearance, the depth D of the groove 3, the width B of the blade tip shoulder arm, the diameter d of the film cooling hole 4, and the rotational speed n of the turbine blade.

[0070] S1.2 Determine the value range and probability distribution form of each uncertain variable.

[0071] The tip clearance height H, the groove depth D, and the tip shoulder width B are all distributed in a Gaussian pattern within their respective ranges, while the diameter d of the film gas hole 4 and the turbine blade rotational speed n are distributed uniformly.

[0072] Table 1. Range and distribution of uncertain variables

[0073]

[0074] Step 2: Based on the five uncertain variables defined in Step 1, construct a five-dimensional parameter space. Using the Latin hypercube sampling method, generate sample sets of different precision in the five-dimensional parameter space and determine the aerothermal performance of the turbine blade tip corresponding to each sample point in the sample set.

[0075] S2.1. Each key structural parameter is treated as a dimension. The five key structural parameters in the uncertainty variables and their corresponding value spaces form a five-dimensional parameter space. The Latin hypercube sampling method is used to generate high-precision and low-precision sample sets in the five-dimensional parameter space.

[0076] In this embodiment, the high-precision sample set includes 17 high-precision sample points, and the low-precision sample set includes 1000 low-precision sample points. The sample points are evenly distributed in adjacent dimensions to cover the parameter fluctuation range.

[0077] S2.2. Based on the sample points in the sample set, establish a three-dimensional model of the turbine blade tip.

[0078] For sample points in both the high-precision and low-precision sample sets, each sample point contains 5 key structural parameters, and a 3D model of the turbine blade tip is constructed based on the sample points.

[0079] S2.3 Perform fluid dynamics simulation calculations on the three-dimensional model of the turbine blade tip to determine the aero-thermal performance of the turbine blade tip, which includes aerodynamic performance and heat transfer performance.

[0080] Aerodynamic performance: Total pressure loss coefficient of turbine engine blades Characterization;

[0081] The aerothermal performance of the turbine blade tip corresponding to the sample point was determined using a steady-state calculation method.

[0082] The total pressure loss coefficient The calculation method is as follows:

[0083]

[0084] In the formula, The relative total pressure at the turbine blade inlet. This is the outlet static pressure of the turbine blades. The relative total pressure at the turbine blade outlet.

[0085] Heat transfer performance: using the area-average heat transfer coefficient of the turbine blade tip The area of ​​the turbine blade tip is characterized by including the blade tip area, the bottom surface of the groove 5, and the sidewall of the groove 6.

[0086] The aero-thermal performance of the turbine blade tip at the sample points was determined using an unsteady gas-thermal coupling calculation method.

[0087] The area-average heat transfer coefficient The calculation method is as follows:

[0088]

[0089] In the formula, For heat flux density, This refers to the total inlet temperature of the turbine blades. This refers to the inlet temperature of the cooling channel for the turbine blades.

[0090] S2.4 Construct a training dataset based on the sample points and their corresponding gas and thermal properties.

[0091] In this embodiment, the sample set includes a high-precision sample set and a low-precision sample set. That is, a second training dataset is constructed based on the high-precision sample set and the corresponding aero-thermal performance of the turbine blade tip. Similarly, a first training dataset is constructed based on the low-precision sample set and the corresponding aero-thermal performance of the turbine blade tip.

[0092] Step 3: Construct a Gaussian process surrogate model based on the sample points generated in Step 2. The Gaussian process surrogate model establishes a stochastic process mapping relationship between the sample points and the gas-thermal performance based on the kernel function. The expected improvement function is used as the acquisition function. New high-precision sample points are generated iteratively through a Bayesian optimization strategy. In each iteration, the gas-thermal performance of the new high-precision sample points is calculated through fluid dynamics simulation and the Gaussian process surrogate model is updated until the Gaussian process surrogate model converges. The newly generated high-precision sample points are then merged into the second training dataset.

[0093] In this embodiment, the kernel function is a radial basis function. Based on the radial basis function, a stochastic process mapping relationship between the input sample points and the gas-thermal performance is established. Then, an expected improvement function is used. The new sample points are determined by the acquisition function. In each iteration, the gas-thermal performance of the new sample points is calculated by fluid dynamics simulation and the Gaussian process surrogate model is updated until the convergence condition of maximum prediction error ≤ 5% is met.

[0094] Expected Improvement Function The expression is as follows:

[0095]

[0096]

[0097] In the formula, Representative sample points The standard deviation of the forecast at that location; Representative sample points The predicted mean at the location; This represents the objective function value of the current best sample. Represents the standard normal cumulative distribution function; This represents the standard normal probability density function. It is dimensionless.

[0098] Step 4: Construct a blade tip gas thermal performance prediction network model using a multi-fidelity deep neural network method, which includes cascaded prediction subnetworks and correction subnetworks.

[0099] The training method for the blade tip gas thermal performance prediction network model is as follows:

[0100] The prediction sub-network is trained based on the first training dataset, and the prediction sub-network outputs the turbine blade tip thermal performance.

[0101] Using the second training dataset and the turbine tip thermal performance output by the prediction sub-network as input, the calibration sub-network is trained. Based on the nonlinear expressive power of the activation function, the mapping and calibration relationship between the turbine tip thermal performance output by the prediction sub-network and the calibration sub-network is determined. The prediction results of the calibration sub-network are iteratively corrected according to the mapping and calibration relationship until the loss function is minimized, and the trained turbine tip thermal performance prediction network model is obtained, and the final prediction result is output.

[0102] The training method for the blade tip gas thermal performance prediction network model specifically includes the following steps:

[0103] S4.1. LF-DNN (Late Fusion Deep Neural Network) is used as the prediction sub-network to learn the global trend of turbine blade tip thermal performance. The prediction sub-network includes an input layer, three hidden layers and an output layer. The number of neurons in the three hidden layers are 64, 64 and 40, respectively.

[0104] The first training dataset is input into the input layer, and the turbine blade tip features are extracted through three hidden layers. The output layer outputs the aero-thermal performance of the turbine blade tip.

[0105] S4.2. A DNN (Deep Neural Network) is used as the calibration sub-network. The calibration sub-network includes an input layer, four hidden layers, and an output layer. The number of neurons in the four hidden layers are 128, 64, 32, and 16, respectively.

[0106] During training, the turbine blade tip thermal performance output from the second training dataset and the prediction sub-network is input into the DNN model for joint training. The nonlinear expressive power of the ReLU (Rectified Linear Unit) activation function is used to accurately learn the mapping and correction relationship between the prediction results of the first and second training datasets. The results of the DNN model are iteratively corrected according to the mapping and correction relationship until the loss function of the correction sub-network is minimized, thereby achieving end-to-end adaptive transfer learning from low precision to high precision. After training, the blade tip thermal performance prediction network model is obtained.

[0107] Loss function of the corrector subnetwork The expression is as follows:

[0108]

[0109] in, For the sample size, To correct the predictions of the subnetwork, For the true value, The regularization coefficient is . As weight, HF To correct the subnetwork, i For the first i 1 sample point.

[0110] The blade tip gas thermal performance prediction network model adopts a two-stage optimization strategy during training. In the first stage, the ADAM (Adaptive Moment Estimation) optimizer is used to achieve fast convergence with a learning rate of 0.001. In the second stage, the L-BFGS (Limited-memory Broyden–Fletcher–Goldfarb–Shanno) optimizer is called to perform high-precision weight fine-tuning.

[0111] Step 5: Sample points are generated by sampling in the multidimensional parameter space of uncertain variables. The air-thermal performance of the sample points is predicted by the blade tip air-thermal performance prediction network model. Then, the probability distribution of air-thermal performance of all sample points is obtained. For air-thermal performance with a probability greater than the set probability, the influence of the corresponding sample points on the air-thermal performance is analyzed.

[0112] S5.1 Perform Monte Carlo sampling in five-dimensional space according to the probability distribution of each parameter;

[0113] For key structural parameters that follow a Gaussian distribution, Box-Muller transformation is used to generate normal random numbers, and for key structural parameters that follow a uniform distribution, inverse transformation is used for sampling. In this embodiment, 100,000 sample points are generated.

[0114] The Box-Muller transform is a mathematical transformation method that converts a uniform distribution into a normal distribution. Box-Muller refers to the surnames of the statisticians George E. Box and Mervin E. Muller, who designed this transformation method.

[0115] S5.2 Input the sample points into the trained blade tip aero-thermal performance prediction network model to obtain the aero-thermal performance of the turbine blade tip. Perform statistical analysis on all aero-thermal performances to determine the distribution of aero-thermal performance.

[0116] Statistical analysis was performed using histograms, kernel density estimation, mean values, and nominal values ​​to analyze the distribution of gas-thermal properties.

[0117] S5.3. Select sample points whose total pressure loss coefficient and area-average heat transfer coefficient change by more than ±3%, and analyze the degree of influence of the sample points on the gas-thermal performance.

[0118] Step 6: Based on the Sobol global sensitivity analysis method, determine the degree of influence of each uncertainty variable on the gas-thermal performance fluctuation in the analysis sample points, where Sobol is a person's name.

[0119] I. The influence of single-source uncertainties on gas-thermal performance is analyzed using the following method:

[0120] 1) Treat one of the key structural parameters in the uncertainty variables as a single design parameter. Other key structural parameters are integrated as constants within the parameter range, thereby eliminating the coupling effects of other key structural parameters;

[0121] 2) Based on the trained blade tip air-thermal performance prediction network model, calculate the weighted average of the air-thermal performance corresponding to each value of a single design parameter, and use the air-thermal performance as the objective function to construct a single design parameter... A single relational model for gas thermal performance ;

[0122]

[0123] in, This represents the mean of the gas thermal properties within the space of uncertain parameter values; For the nth parameter;

[0124] 3) Calculate single design parameters based on the single relation model. corresponding variance With total variance The proportion is used to obtain the degree of influence of a single design parameter on gas thermal performance.

[0125] The variance calculation method for a single design parameter is as follows:

[0126]

[0127] in, For single design parameters Independent contributions to gas-thermal performance Indicates the integral;

[0128] 4) Repeat the above process to determine the degree of influence of each key structural parameter on the gas-thermal performance, and rank the degree of influence according to the ratio of variance to total variance.

[0129] II. The influence of two sources of uncertainty on gas-thermal performance is analyzed using the following method:

[0130] 10) Two key structural parameters in the uncertainty variables ( As a dual design parameter, other key structural parameters are integrated as constants within the parameter range, thereby eliminating the coupling effects of other key structural parameters;

[0131] 20) Based on the trained blade tip air-thermal performance prediction network model, calculate the weighted average of the air-thermal performance corresponding to each value of the dual design parameters. Using the air-thermal performance as the objective function, construct an interaction model of the dual design parameters on the air-thermal performance. This interaction model is a binary function. .

[0132] The expression for this interaction relationship model is as follows:

[0133]

[0134] in, Key structural parameters Independent contributions to gas-thermal performance Key structural parameters Independent contribution to gas thermal performance.

[0135] 30) Calculate the ratio of the coupling variance of the two design parameters to the total variance based on the interaction model, and then obtain the degree of influence of the interaction of the two design parameters on the gas-thermal performance.

[0136] The expression for the coupling variance is as follows:

[0137]

[0138] in, This is the coupling contribution term of the two parameters to the gas-thermal performance;

[0139] The expression for the total variance is as follows:

[0140]

[0141] Step 7: Repeat step 6 to obtain the ratio of the variance of each design parameter to the total variance. Sort the degree of influence according to the ratio, filter significant parameters according to the ratio threshold, determine the optimal value of significant parameters according to the single relation model, and construct the parameter combination that optimizes the gas-thermal performance based on the optimal value.

[0142] The degree of influence was ranked according to proportion, that is, the effects of tip clearance height H, groove depth D, tip shoulder width B, film cooling hole diameter d, and turbine blade rotational speed n on aerothermal performance were identified and screened. The parameters are significant parameters. The optimal values ​​of the significant parameters are determined based on the single relation model, and finally the parameter combination that optimizes the gas-thermal performance can be obtained.

[0143] This uncertainty analysis method for turbine tip aero-thermal performance overcomes the dimensionality limitations of traditional high-dimensional uncertainty analysis methods by constructing a cascaded architecture of multi-fidelity deep neural networks and combining ReLU activation function with Bayesian optimization strategy. It significantly improves computational efficiency by 94.5% while raising prediction accuracy to 98.2% (error ≤ 3.8%), achieving adaptive modeling of aero-thermal performance for input variables with arbitrary probability distributions. This method provides a high-precision, low-cost, robust optimization decision-making basis for aero-engine turbine tip design, effectively supporting the efficient and stable design of high-load turbine tips.

[0144] Example 2

[0145] Correspondingly, this application also provides an uncertainty analysis system for the thermal performance of turbine blade tip gas, including:

[0146] The acquisition module is used to determine the key structural parameters of the turbine blade tip and use them as uncertainty variables. Based on the uncertainty variables, a sample set containing sample points with different precision is generated, the aero-thermal performance of each sample point is determined, and then a first training dataset and a second training dataset with different precision are constructed.

[0147] The training module is used to construct a prediction network model for the thermal performance of the blade tip using cascaded prediction and correction subnetworks.

[0148] The prediction subnetwork is trained using the first training dataset to learn the global trend of air-thermal performance. Then, the correction subnetwork is trained together with the second training dataset and the output of the prediction subnetwork to establish a mapping correction relationship from low precision to high precision until the loss function is minimized, thus obtaining the trained blade tip air-thermal performance prediction network model.

[0149] The statistical module is used to generate analysis sample points based on the uncertain variables, use the trained blade tip air-thermal performance prediction network model to predict air-thermal performance, obtain the air-thermal performance probability distribution through statistical analysis of the prediction results, and screen out analysis sample points whose air-thermal performance probability exceeds a preset threshold.

[0150] The analysis module is used to determine the degree of influence of each uncertainty variable on the fluctuation of gas-thermal performance in the analysis sample points based on the Sobol global sensitivity analysis method.

[0151] The optimization module is used to sort all uncertain variables by their degree of influence, filter out significant parameters based on preset thresholds, determine the optimal values ​​of each significant parameter, and combine them to form a parameter combination that makes the turbine tip air thermal performance approach the global optimum.

[0152] It should be noted that, in the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of each module is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another device, or some features may be ignored or not executed. The modules described as separate components may or may not be physically separated. The components shown as modules may be one or more physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs.

[0153] Furthermore, in the various embodiments of the present invention, the modules can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The integrated unit described above can be implemented in hardware or as a software functional unit.

[0154] An electronic device provided in this application includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps of an uncertainty analysis method for turbine blade tip gas thermal performance as described in any of the above embodiments.

[0155] Another electronic device provided in this application embodiment may further include: an input port connected to a processor for transmitting data collected by an external acquisition device to the processor; a display unit connected to the processor for displaying the processor's processing results to the outside world; and a communication module connected to the processor for enabling communication between the electronic device and the outside world. The display unit may be a display panel, a laser scanning display, etc.; the communication method adopted by the communication module includes, but is not limited to, mobile high-definition link technology, universal serial bus, high-definition multimedia interface, and wireless connection (including wireless fidelity technology, Bluetooth communication technology, and Bluetooth low power communication technology).

[0156] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the steps of an uncertainty analysis method for turbine blade tip gas thermal performance as described in any of the above embodiments.

[0157] For descriptions of relevant parts in the uncertainty analysis system, electronic device, and computer-readable storage medium for turbine blade tip gas thermal performance provided in this application, please refer to the detailed description of the corresponding parts in the uncertainty analysis method for turbine blade tip gas thermal performance provided in this application, which will not be repeated here. Furthermore, parts of the technical solutions provided in this application that are consistent with the implementation principles of corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0158] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for uncertainty analysis of the thermal properties of turbine blade tip gas, characterized in that, Includes the following steps: Step 1: Determine the key structural parameters of the turbine blade tip and use them as uncertainty variables. Based on the uncertainty variables, generate a sample set containing sample points with different precision, determine the aero-thermal performance of each sample point, and then construct a first training dataset and a second training dataset with different precision. Step 2: Construct a blade tip gas thermal performance prediction network model using cascaded prediction sub-networks and correction sub-networks; The prediction subnetwork is trained using the first training dataset to learn the global trend of air-thermal performance. Then, the correction subnetwork is trained together with the second training dataset and the output of the prediction subnetwork to establish a mapping correction relationship from low precision to high precision until the loss function is minimized, thus obtaining the trained blade tip air-thermal performance prediction network model. Step 3: Generate analysis sample points based on the uncertain variables, use the trained blade tip air-thermal performance prediction network model to predict air-thermal performance, obtain the air-thermal performance probability distribution through statistical analysis of the prediction results, and screen out analysis sample points whose air-thermal performance probability exceeds the preset threshold. The generation of analysis sample points based on the uncertainty variables includes: Monte Carlo sampling is performed in the multidimensional parameter space composed of uncertain variables according to the probability distribution of each uncertain variable. For key structural parameters that follow a Gaussian distribution, Box-Muller transformation is used to generate normal random numbers, and for key structural parameters that follow a uniform distribution, inverse transformation is used to sample and generate analysis sample points. Step 4: Based on the Sobol global sensitivity analysis method, determine the degree of influence of each uncertainty variable on the fluctuation of gas-thermal performance in the analysis sample points; Step 5: Sort all uncertain variables by their degree of influence, select significant parameters based on preset thresholds, determine the optimal values ​​of each significant parameter, and combine them to form a parameter combination that makes the turbine tip gas thermal performance approach the global optimum.

2. The uncertainty analysis method for the thermal performance of turbine blade tip gas according to claim 1, characterized in that, In step 1, determining the gas-thermal properties of each sample point specifically involves: A three-dimensional model of the turbine blade tip corresponding to each sample point is established. The gas-thermal performance of the sample points is obtained by simulating the three-dimensional model of the turbine blade tip through fluid dynamics. The gas-thermal performance includes the total pressure loss coefficient and the area-average heat transfer coefficient.

3. The uncertainty analysis method for the thermal performance of turbine blade tip gas according to claim 1, characterized in that, Before step 2, the following are also included: A Gaussian process surrogate model is constructed based on the sample points in the second training dataset. A stochastic process mapping relationship between the sample points and the gas-thermal performance is established. The expected improvement function is used as the acquisition function. New high-precision sample points are generated iteratively through a Bayesian optimization strategy. The gas-thermal performance of the new high-precision sample points is calculated by simulation and the Gaussian process surrogate model is updated until the Gaussian process surrogate model converges. The new high-precision sample points are then merged into the second training dataset.

4. The uncertainty analysis method for the thermal performance of turbine blade tip gas according to claim 3, characterized in that, The process of establishing a mapping relationship between sample points and gas thermal properties includes: A stochastic process mapping relationship between input sample points and gas-thermal properties is established based on radial basis functions.

5. The uncertainty analysis method for the thermal performance of turbine blade tip gas according to claim 1, characterized in that, The Sobol-based global sensitivity analysis method determines the degree of influence of each uncertainty variable on the fluctuation of gas-thermal performance in the analysis sample points, including: One key structural parameter among the uncertain variables is treated as a single design parameter, while the other key structural parameters are integrated into a constant within their parameter range. Based on the trained blade tip air-thermal performance prediction network model, the weighted average value of the air-thermal performance corresponding to each value of a single design parameter is calculated. Using the air-thermal performance as the objective function, a single relationship model between a single design parameter and air-thermal performance is constructed. The ratio of the variance corresponding to a single design parameter to the total variance is calculated based on the single-relationship model to obtain the degree of influence of the single design parameter on the gas-thermal performance.

6. The uncertainty analysis method for the thermal performance of turbine blade tip gas according to claim 5, characterized in that, It also includes an analysis of the interactive effects of multiple design parameters on gas-thermal performance: Multiple key structural parameters among the uncertain variables are treated as multiple design parameters, and other key structural parameters are integrated into constants within their parameter ranges. Based on the trained blade tip air-thermal performance prediction network model, the weighted average of the air-thermal performance corresponding to each value of multiple design parameters is calculated. Using air-thermal performance as the objective function, an interaction model of multiple design parameters on air-thermal performance is constructed. The ratio of the coupling variance to the total variance of multiple design parameters is calculated based on the interaction model, thus obtaining the degree of influence of the interaction of multiple design parameters on gas-thermal performance.

7. The uncertainty analysis method for the thermal performance of turbine blade tip gas according to claim 6, characterized in that, Determining the optimal values ​​of each significant parameter includes: Based on the Sobol global sensitivity analysis method, the ratio of the variance of each design parameter to the total variance is determined. The influence is ranked according to the ratio, significant parameters are screened according to the ratio threshold, the optimal value of the significant parameters is determined according to the single relation model, and the optimal parameter combination is constructed according to the optimal value.

8. An uncertainty analysis system for the thermal performance of turbine blade tip gas, characterized in that, include: The acquisition module is used to determine the key structural parameters of the turbine blade tip and use them as uncertainty variables. Based on the uncertainty variables, a sample set containing sample points with different precision is generated, the aero-thermal performance of each sample point is determined, and then a first training dataset and a second training dataset with different precision are constructed. The training module is used to construct a prediction network model for the thermal performance of the blade tip using cascaded prediction and correction subnetworks. The prediction subnetwork is trained using the first training dataset to learn the global trend of air-thermal performance. Then, the correction subnetwork is trained together with the second training dataset and the output of the prediction subnetwork to establish a mapping correction relationship from low precision to high precision until the loss function is minimized, thus obtaining the trained blade tip air-thermal performance prediction network model. The statistical module is used to generate analysis sample points based on the uncertain variables, use the trained blade tip air-thermal performance prediction network model to predict air-thermal performance, obtain the air-thermal performance probability distribution through statistical analysis of the prediction results, and screen out analysis sample points whose air-thermal performance probability exceeds a preset threshold. The generation of analysis sample points based on the uncertainty variables includes: Monte Carlo sampling is performed in the multidimensional parameter space composed of uncertain variables according to the probability distribution of each uncertain variable. For key structural parameters that follow a Gaussian distribution, Box-Muller transformation is used to generate normal random numbers, and for key structural parameters that follow a uniform distribution, inverse transformation is used to sample and generate analysis sample points. The analysis module is used to determine the degree of influence of each uncertainty variable on the fluctuation of gas-thermal performance in the analysis sample points based on the Sobol global sensitivity analysis method. The optimization module is used to sort all uncertain variables by their degree of influence, filter out significant parameters based on preset thresholds, determine the optimal values ​​of each significant parameter, and combine them to form a parameter combination that makes the turbine tip air thermal performance approach the global optimum.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the uncertainty analysis method for the thermal performance of turbine blade tip gas as described in any one of claims 1 to 7.

Images