CN-122020915-A - Uncertainty analysis method and system for turbine blade tip aerothermal performance

CN122020915ACN 122020915 ACN122020915 ACN 122020915ACN-122020915-A

Abstract

The invention discloses an uncertainty analysis method and system for gas-heat performance of a turbine blade top, wherein key structural parameters of the turbine blade top are firstly determined to serve as uncertainty variables, a multi-precision sample set is built, a multi-fidelity deep neural network model is built by adopting a cascade prediction sub-network and a cascade correction sub-network, a mapping correction relation from low precision to high precision is built through staged training, further, the trained blade top gas-heat performance prediction network model is utilized to conduct gas-heat performance distribution analysis, the influence degree of each uncertainty variable on performance fluctuation is quantified by combining with a Sobol global sensitivity analysis method, and finally significant parameters are sequenced and screened according to the influence degree, and optimal value combinations of the significant parameters are determined to form a design scheme for enabling the gas-heat performance of the turbine blade top to approach global optimum. The method effectively breaks through the dimension limitation of high-dimensional uncertainty analysis, remarkably improves the calculation efficiency and the prediction precision, and provides reliable technical support for the robust design of the turbine blade tip.

Inventors

LI CHENXI
Huang Fangzhou
PAN JIAPING
GAO XU

Assignees

西安理工大学

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The uncertainty analysis method for the gas-thermal performance of the turbine blade tip is characterized by comprising the following steps of: Step 1, determining key structural parameters of a turbine blade top, and taking the key structural parameters as uncertainty variables, generating sample sets containing sample points with different precision based on the uncertainty variables, determining the gas-heat performance of each sample point, and further constructing a first training data set and a second training data set with different precision; Step 2, constructing a leaf top gas-heat performance prediction network model by adopting a cascade prediction sub-network and a corrector sub-network; Training a prediction sub-network by adopting a first training data set to learn the global trend of the gas-heat performance, and then combining a second training data set and the output of the prediction sub-network to train a correction sub-network together, so as to establish a mapping correction relation from low precision to high precision until a loss function is minimized, thereby obtaining a trained leaf top gas-heat performance prediction network model; Step 3, generating analysis sample points based on the uncertainty variable, carrying out gas-heat performance prediction by using a trained leaf top gas-heat performance prediction network model, obtaining gas-heat performance probability distribution by carrying out statistical analysis on a prediction result, and screening the analysis sample points of which the gas-heat performance probability exceeds a preset threshold; step 4, determining the influence degree of each uncertainty variable in the analysis sample point on the gas-heat performance fluctuation based on a Sobol global sensitivity analysis method; And 5, sequencing the influence degree of all uncertainty variables, screening out significant parameters according to a preset threshold value, determining the optimal value of each significant parameter, and combining to form a parameter combination which enables the gas-thermal performance of the turbine blade top to approach global optimal.
2. The method for uncertainty analysis of turbine blade tip aerothermal properties according to claim 1, wherein in step 1, the determining aerothermal properties of each sample point is specifically: and establishing a turbine blade top three-dimensional model corresponding to each sample point, and obtaining the gas-heat performance of the sample points by simulating the turbine blade top three-dimensional model through fluid dynamics, wherein the gas-heat performance comprises a total pressure loss coefficient and an area average heat exchange coefficient.
3. The method for uncertainty analysis of turbine blade tip aerothermal properties according to claim 1, further comprising, prior to step2: And constructing a Gaussian process proxy model based on the sample points in the second training data set, establishing a random process mapping relation of the sample points and the gas-heat performance, taking an expected improvement function as an acquisition function, iteratively generating a new high-precision sample point through a Bayesian optimization strategy, calculating the gas-heat performance of the new high-precision sample point through simulation and updating the Gaussian process proxy model until the Gaussian process proxy model converges, and merging the new high-precision sample point into the second training data set.
4. A method for uncertainty analysis of gas-thermal properties of a turbine blade tip according to claim 3, wherein said establishing a random process mapping relationship between sample points and gas-thermal properties comprises: and establishing a random process mapping relation between the input sample points and the gas-heat performance according to the radial basis function.
5. The method for uncertainty analysis of turbine blade tip aero-thermal properties of claim 1, wherein said generating analysis sample points based on said uncertainty variables comprises: In a multidimensional parameter space formed by uncertainty variables, performing Monte Carlo sampling in the multidimensional parameter space according to probability distribution of each uncertainty variable; And generating normal random numbers by using Box-Muller transformation on the key structure parameters obeying Gaussian distribution, and generating analysis sample points by using inverse transformation sampling on the key structure parameters obeying uniform distribution.
6. The method for analyzing the uncertainty of the gas-thermal performance of the turbine blade tip according to claim 1, wherein the determining the influence degree of each uncertainty variable in the analysis sample point on the gas-thermal performance fluctuation based on the Sobol global sensitivity analysis method comprises the following steps: taking one key structural parameter in the uncertainty variable as a single design parameter, and integrating other key structural parameters into a constant in the parameter range; based on the trained leaf top gas thermal performance prediction network model, calculating a weighted average value of the gas thermal performance corresponding to each value of the single design parameter, and constructing a single relation model of the single design parameter to the gas thermal performance by taking the gas thermal performance as an objective function; and calculating the ratio of the variance corresponding to the single design parameter to the total variance according to the single relation model to obtain the influence degree of the single design parameter on the gas-thermal performance.
7. The method for uncertainty analysis of turbine blade tip aero-thermal properties according to claim 6, further comprising an interactive impact analysis of multiple design parameters on aero-thermal properties: Taking a plurality of key structural parameters in the uncertainty variable as a plurality of design parameters, and integrating other key structural parameters into constants in the parameter range; based on the trained leaf top gas thermal performance prediction network model, calculating a weighted average value of the gas thermal performance corresponding to each value of the multiple design parameters, and constructing an interaction relation model of the multiple design parameters on the gas thermal performance by taking the gas thermal performance as an objective function; And calculating the ratio of the coupling variance to the total variance of the multiple design parameters according to the interaction relation model to obtain the influence degree of the interaction of the multiple design parameters on the gas-heat performance.
8. The method for uncertainty analysis of turbine blade tip aerothermal properties of claim 6, wherein said determining optimal values for each salient parameter comprises: Based on a Sobol global sensitivity analysis method, determining the proportion of the variance to the total variance of each design parameter, sorting the influence degree according to the proportion, screening the significant parameters according to a proportion threshold, determining the optimal value of the significant parameters according to a single relation model, and constructing an optimal parameter combination according to the optimal value.
9. An uncertainty analysis system for turbine blade tip aerothermal performance, comprising: the acquisition module is used for determining key structural parameters of the turbine blade tip and serving as uncertainty variables, generating sample sets containing sample points with different precision based on the uncertainty variables, determining the gas-heat performance of each sample point, and further constructing a first training data set and a second training data set with different precision; the training module is used for constructing a leaf top gas-heat performance prediction network model by adopting a cascaded prediction sub-network and a cascaded correction sub-network; Training a prediction sub-network by adopting a first training data set to learn the global trend of the gas-heat performance, and then combining a second training data set and the output of the prediction sub-network to train a correction sub-network together, so as to establish a mapping correction relation from low precision to high precision until a loss function is minimized, thereby obtaining a trained leaf top gas-heat performance prediction network model; The statistical module is used for generating analysis sample points based on the uncertainty variable, carrying out gas-heat performance prediction by utilizing a trained leaf top gas-heat performance prediction network model, obtaining gas-heat performance probability distribution through statistical analysis of a prediction result, and screening the analysis sample points of which the gas-heat performance probability exceeds a preset threshold; The analysis module is used for determining the influence degree of each uncertainty variable in the analysis sample point on the gas-heat performance fluctuation based on a Sobol global sensitivity analysis method; The optimization module is used for sequencing the influence degree of all uncertainty variables, screening out significant parameters according to a preset threshold value, determining the optimal value of each significant parameter, and combining to form a parameter combination which enables the gas-thermal performance of the turbine blade top to approach global optimum.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, carries out the steps of a method for uncertainty analysis of aero-thermal properties of a turbine blade tip according to any one of claims 1 to 8.

Description

Uncertainty analysis method and system for turbine blade tip aerothermal performance Technical Field The invention relates to the technical field of turbine aerodynamic performance analysis, in particular to an uncertainty analysis method for the aerodynamic performance of a high-load turbine blade tip. Background The power devices such as the gas turbine, the aeroengine and the like have compact overall structure, the geometric dimension of the turbine blade tip of the core component is especially tiny, the processing and manufacturing difficulty is high, and the tiny dimensional deviation can cause the obvious change of the shape of the turbine blade tip. Meanwhile, the transonic turbine blade tip directly bears high-temperature high-pressure gas impact in a high-speed rotation state, and the abrupt change of working condition parameters in the starting and stopping processes can cause the fluctuation of the working state of the turbine blade tip, so that the turbine blade tip deviates from ideal design conditions. Such uncertainty factors can significantly affect the aero-thermal properties of the turbine blade tips. Therefore, in the design process of the turbine blade tip, the uncertainty of the geometric form and the operation working condition is comprehensively considered, and the influence rule of the geometric form and the operation working condition on the gas-heat performance of the turbine blade tip is accurately predicted, so that the safe and stable operation of the gas turbine is reliably ensured. With the continuous development of uncertainty quantization methods, the existing multi-fidelity method (such as a polynomial chaotic expansion method) realizes the mapping relation between uncertainty input and statistical characteristics by constructing an orthogonal polynomial basis. However, its core limitation is that the dependence on input distribution is strong, and the problem of dimension disaster is easily encountered when facing high-dimension uncertainty analysis. The dimension disaster has become a key bottleneck for restricting the popularization and application of the traditional method in engineering practice. Taking the turbine blade top gas-heat performance analysis scene as an example, when the dimension of an uncertain variable is obviously increased, the number of high-precision samples required by the polynomial chaotic expansion method is exponentially increased, so that the demand for computing resources is rapidly increased, and particularly in the analysis involving a high-dimensional parameter space (such as more than 15 dimensions), in order to maintain the prediction precision, simulation resources with far-beyond engineering actual bearing capacity are required to be called, so that a contradiction which is difficult to reconcile between the computing efficiency and the model precision is formed. Therefore, how to break through the resource constraint in the high-dimensional uncertainty quantification and construct an analysis method with both high efficiency and reliability has become a key problem to be solved in the current engineering optimization field. Disclosure of Invention Aiming at the problems in the prior art, the invention provides the uncertainty analysis method and the system for the gas-heat performance of the turbine blade tip, which can effectively break through the dimension limitation of uncertainty analysis, obviously improve the calculation efficiency, improve the uncertainty quantification precision of the high-load turbine blade tip to more than 98 percent and provide a powerful support for the efficient and stable design of an aeroengine. The invention is realized by the following technical scheme: In a first aspect, the application provides a method for analyzing uncertainty of aerothermal properties of a turbine blade tip, comprising the following steps: Step 1, determining key structural parameters of a turbine blade top, and taking the key structural parameters as uncertainty variables, generating sample sets containing sample points with different precision based on the uncertainty variables, determining the gas-heat performance of each sample point, and further constructing a first training data set and a second training data set with different precision; Step 2, constructing a leaf top gas-heat performance prediction network model by adopting a cascade prediction sub-network and a corrector sub-network; Training a prediction sub-network by adopting a first training data set to learn the global trend of the gas-heat performance, and then combining a second training data set and the output of the prediction sub-network to train a correction sub-network together, so as to establish a mapping correction relation from low precision to high precision until a loss function is minimized, thereby obtaining a trained leaf top gas-heat performance prediction network model; Step 3, generating analysis sample points based on the uncertainty variable