Engine turbine blade digital twin model modeling method based on data driving

By adopting a data-driven method in turbine blade modeling, combining optimized sampling and chaotic polynomial expansion technology, a random matrix mapping model of low-dimensional popular features and full-field performance response is constructed, solving the multi-source uncertainty in the existing technology It deals with difficult problems and large-scale sample data requirements, and achieves efficient and accurate full-field performance prediction and optimized design.

CN119004691AActive Publication Date: 2024-11-22HUNAN UNIV

Patent Information

Application Number
CN202410946719.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-22
Estimated Expiration
2044-07-16

AI Technical Summary

Technical Problem

The existing digital twin technology is difficult to effectively deal with complex multi-source uncertainty in turbine blade modeling, large sample data demands and high computational costs, making it difficult to accurately predict the full-field performance response, especially in extreme working environments.

Method used

Using a data-driven method, a finite element model of thermal coupled turbine blades is established, and the optimal distributed sample points are obtained through optimized sampling strategies. Machine learning and chaotic polynomial expansion technology are used to perform dimensionality reduction and uncertainty analysis, and low-dimensional popular features and full A random matrix mapping model of field performance responses enables rapid prediction and visualization of multi-physics performance.

Benefits of technology

It significantly improves the accuracy and reliability of turbine blade performance prediction, simplifies modeling steps, reduces calculation costs, and can achieve efficient full-field performance response prediction under limited sample data, providing a more scientific basis for optimized design.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004691A_ABST
    Figure CN119004691A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of aero-engine design and simulation, in particular to an engine turbine blade digital twin model modeling method based on data driving. According to the engine turbine blade digital twin modeling method based on data driving, high-dimensional response data are obtained by combining a small amount of sampling with physical model deviation; constructing a model parameter to a full-field response popular structure through popular learning dimensionality reduction; uncertainties are analyzed by adopting chaos polynomial expansion, and uncertainty mapping from parameters to popular characteristics is established; and a random moment mapping model is constructed, and rapid prediction of multi-physical field performance probability behaviors is realized. The fluctuation characteristics are quantized and visualized, a non-intrusive digital twinning framework is established, and the multi-field coupling problem is flexibly solved; physical test dependence is reduced, and a solid guarantee is provided for multi-working-condition performance prediction and optimization of the turbine blade.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of aero-engine design and simulation, and in particular to a data-driven digital twin modeling method for engine turbine blades. Background Art

[0002] As the core component of an aircraft engine, the stability and reliability of the performance of the engine turbine blades have a crucial impact on the operating efficiency and safety of the entire machine. Turbine blades are subjected to complex mechanical and thermodynamic loads in extreme working environments (high temperature, high pressure, and high-speed rotation). These loads include thermal stress and mechanical stress generated by gas flow, which require the blades to maintain extremely high structural integrity and durability. However, due to the existence of various uncertain factors such as material properties, manufacturing processes, and operating conditions, the actual performance of turbine blades often deviates from design expectations, which limits the optimization of engine performance.

[0003] Traditional turbine blade design and analysis methods mainly rely on deterministic computational models, which have limitations in dealing with complex physical phenomena and multi-source uncertainties, and are difficult to fully capture and quantify performance fluctuations under actual working conditions. With the rapid development of computer technology and simulation science, digital twin technology, as an emerging modeling and analysis method, achieves accurate mapping and real-time interaction between physical space and digital space by building virtual mirrors of physical objects, providing a new way for performance prediction and optimization design of turbine blades.

[0004] However, existing digital twin technology still faces many challenges when applied to turbine blade modeling:

[0005] Complexity of application scenarios: The working environment of turbine blades is extreme and changeable, involving multidisciplinary physical processes such as thermodynamics, fluid mechanics, solid mechanics, etc., which places extremely high demands on the accuracy and robustness of the digital twin model.

[0006] The cumbersomeness of modeling steps and details: Building an accurate digital twin model of turbine blades requires detailed modeling steps and precise parameter settings, including the establishment of a thermal-mechanical coupling finite element model, the determination of the model parameter deviation model, the collection and processing of sample data, etc. The errors in each step may accumulate and affect the prediction accuracy of the final model.

[0007] Limitations of the random moment mapping model: In existing digital twin technologies, probabilistic behavior prediction often relies on a large amount of sample data for training and verification, which not only increases the computational cost, but may also lead to insufficient generalization of the model when data is scarce or unevenly distributed. In addition, traditional methods are inefficient in dealing with uncertainty quantification of high-dimensional response fields, making it difficult to quickly and accurately predict the uncertainty behavior of the full-field performance response.

[0008] Accuracy issues in full-field performance response prediction: The full-field performance response prediction of turbine blades requires comprehensive consideration of the interactions of multiple physical fields and multiple physical quantities. Existing models often have insufficient accuracy when dealing with such complex problems, and it is difficult to accurately reflect the probabilistic characteristics of performance fluctuations and variations under actual operating conditions. Summary of the invention

[0009] In response to the above problems, the present invention aims to develop a digital twin modeling method that can efficiently handle multi-source uncertainties, has low requirements for sample data, and can accurately predict the full-field performance response of turbine blades. It has important engineering value and economic significance for improving the accuracy and reliability of engine design.

[0010] The technical solution adopted by the present invention is: a data-driven digital twin modeling method for engine turbine blades, comprising the following steps:

[0011] Step 1: Establish a thermal-mechanical coupling finite element model of the turbine blade, and determine the deviation model of the blade structural parameters based on engineering experience or experimental data, wherein the model parameters include elastic modulus, Poisson's ratio, pressure side load, suction side load, thermal expansion coefficient, thermal conductivity, internal cooling convection coefficient, suction side convection coefficient, internal cooling ambient temperature, and external blade ambient temperature;

[0012] Step 2: Based on the deviation model and finite element analysis, an optimized sampling strategy is used to obtain the optimal distribution sample points of the model parameters in the [0,1] cumulative probability space, as well as the corresponding full-field performance response data;

[0013] Step 3: Use machine learning methods to perform popular dimension reduction on high-dimensional full-field response data, extract low-dimensional features and corresponding popular functions according to the total empirical variance criterion, and then construct a popular structure from model parameters to full-field performance response;

[0014] Step 4: Perform uncertainty propagation analysis on the features after dimensionality reduction through the chaotic polynomial expansion method, solve the chaotic polynomial model coefficients to estimate the uncertainty of low-dimensional epidemic features, and establish an uncertainty mapping model from model parameters to epidemic features;

[0015] Step 5: Based on the epidemic structure, a random moment mapping model of low-dimensional epidemic characteristics and the full-field performance response of blade thermodynamics is established to achieve rapid prediction of the probability characteristics of blade multi-physical field performance under multi-source deviations of model parameters;

[0016] Step 6: The sampling mean, variance, skewness and kurtosis are used to quantify the probability characteristics of the full-field response, and the maximum entropy principle is used to visualize the probability states such as deviation or fluctuation of the blade physical performance.

[0017] The deviation model in step 1 uses a probability distribution model to characterize the uncertainty of each model parameter.

[0018] The step 2 first obtains the sample data z with the best distribution in the [0,1] cumulative probability space according to the following minimization criterion: (i) ,i=1,2,…,s

[0019]

[0020] where d j For any two points z (i) and z (j) The ascending set of distances (d1, d2, ..., d j ,…,d η ) in the jth distance, J j Indicates that the distance between all pairs of sample points is exactly equal to d j The optimal sample data in the parameter space can be obtained by the following formula:

[0021] x (i) =F -1 (z (i) ) (9)

[0022] Where F -1 is a probability distribution optimization function based on parameter x. Based on the above sampling strategy, the model parameter x=[x1,x2,…x n ] Corresponding sample data Then, finite element simulations are performed to obtain the corresponding full-field performance response data Y = [y1, y2, ..., y s ],y i =[y1,y2,…,y m ] T .

[0023] The step 3 is performed as follows:

[0024]

[0025] In the formula By solving equation (3), the corresponding popular features and feature vectors can be solved. The sorted feature values ​​are λ1≥λ2≥…≥λ m , the corresponding eigenvector is V = [v1,v2,…,v m ]. According to the total empirical variance criterion, The number of effective low-dimensional features can be determined. For the retained m'-dimensional low-dimensional features, its popular structure can be expressed as

[0026]

[0027] Based on the above popular structure, the large-dimensional data in the original space can be compressed into small-dimensional data in the manifold space, retaining most of the information in the large data set. At the same time, the high-dimensional response can also be reconstructed by superimposing the information contained in the m'-dimensional low-dimensional features through this structure.

[0028] In step 4, the first second-order origin moment of the popular feature is obtained by the following formula

[0029]

[0030] Where β α (α=0,1,2,…,P-1) is the coefficient of the chaotic polynomial model of the t-th low-dimensional feature established, which can be obtained by minimizing the following L2 norm residual

[0031]

[0032] In the formula, Represents the i-th high-dimensional response y i =[y1,y2,…,y m ] T The low-dimensional feature response after compression by the popular structure. and According to formula (6), using the response and To construct the corresponding chaotic polynomial and get.

[0033] The step 5 is to establish a random moment mapping model of low-dimensional popular features and full-field performance response according to the following formula:

[0034]

[0035] In the formula is the quadratic expansion coefficient, k = 1, 2, 3, 4. Based on this, a random moment mapping model of low-dimensional popular features and full-field performance responses is established. The first four-order origin moments of the high-dimensional responses of each dimension can be obtained through the first four-order origin moments of the low-dimensional feature responses.

[0036] The step 1 is to visualize the uncertainty characteristics such as uncertainty fluctuation and variation of the full-field response as follows:

[0037]

[0038] Based on the above random moments, the maximum entropy principle can be used to solve the probability density function of the response and realize the visualization of its probability distribution characteristics.

[0039] The method further includes an adaptive adjustment step for the complexity of the application scenario, by real-time monitoring of changes in the working environment parameters of the turbine blades, and dynamically adjusting the boundary conditions and material property parameters in the thermal-mechanical coupling finite element model to achieve accurate simulation of the uncertain performance of the turbine blades under different working conditions.

[0040] The method introduces automated process control in the modeling steps, automatically executes steps 1 to 6 through preset scripts or software tools, reduces human intervention, improves modeling efficiency and consistency, and ensures the accuracy and repeatability of modeling details.

[0041] The random moment mapping model iteratively optimizes the mapping relationship between low-dimensional popular features and full-field performance responses by introducing high-order chaotic polynomial expansion and reinforcement learning algorithm, thereby improving the accuracy and generalization ability of the random moment mapping model and achieving high-precision simulation prediction of the probabilistic behavior of full-field performance responses.

[0042] The beneficial effects of the present invention are as follows: the present invention innovatively proposes a digital twin model construction technology for engine turbine blades, which can systematically quantify the global effects of key factors such as environment, structure and materials on blade performance, and realize comprehensive prediction, in-depth analysis and precise understanding of the mechanical behavior of blades under arbitrary inputs and their deviation fluctuations; at the same time, the present invention constructs a data-driven non-invasive digital twin model framework, which only requires limited sample data and can efficiently predict the uncertainty fluctuations and variation characteristics of structural behavior without touching the existing structural system, providing a flexible and powerful framework for structural performance evaluation and optimization design in complex multi-field coupling environments. A large modeling and analysis tool; Furthermore, the digital twin model construction strategy adopted by the present invention cleverly combines manifold learning and polynomial chaos expansion technology, successfully transforms the uncertainty quantification problem of high-dimensional structural response field into efficient analysis of low-dimensional feature space, and constructs an accurate mapping model from model parameters to random moments of high-dimensional response field, intuitively presents the probability distribution characteristics of mechanical behavior at any position, significantly improves the efficiency and accuracy of model construction, and shows significant beneficial effects in many aspects, which not only improves the accuracy of the prediction of the probability characteristics of turbine blade performance, but also optimizes the modeling process and reduces the computing cost, which is specifically manifested in the following aspects:

[0043] (1) Comprehensively evaluate the global impact of key factors on blade performance: By constructing a digital twin model of the engine turbine blade, the present invention can systematically consider the comprehensive impact of multiple key factors such as environment, structure, and materials on blade performance. Compared with traditional methods, the present invention is no longer limited to the analysis of a single or a few parameters, but comprehensively integrates these complex factors to achieve a global quantitative evaluation of blade structural performance. This feature enables the present invention to have higher accuracy and comprehensiveness in predicting, analyzing, and understanding the mechanical behavior of blades under different input conditions and their deviation fluctuations, providing a more scientific basis for the optimal design of turbine blades.

[0044] (2) Effectively solve the complexity of application scenarios: In view of the extreme and changeable working environment of turbine blades and the complexity of multidisciplinary cross-disciplinary physical processes, the present invention integrates knowledge from multiple disciplines such as thermodynamics, fluid mechanics, and solid mechanics by constructing a comprehensive digital twin model to achieve accurate simulation of complex physical phenomena. This method not only takes into account the thermal stress and mechanical stress of the blades under high temperature, high pressure, and high-speed rotation, but also comprehensively considers various uncertain factors such as material properties, manufacturing processes, and operating conditions, so as to more accurately reflect the performance of the blades under actual working conditions. This comprehensive simulation capability provides strong support for the performance evaluation and optimal design of turbine blades in complex application scenarios.

[0045] (3) Simplify modeling steps and details, and improve modeling efficiency: The construction process of the traditional turbine blade digital twin model is cumbersome and complicated, involving a large number of modeling steps and sophisticated parameter settings. The present invention significantly simplifies the modeling process and reduces the requirements for modeling details by introducing optimized sampling strategies, machine learning methods, and advanced algorithms such as manifold learning and polynomial chaos expansion. Specifically, the optimized sampling strategy can obtain the most representative initial sample points under limited sample data and reduce the number of simulations; the machine learning method can automatically extract key features from high-dimensional data and reduce the complexity of data processing; manifold learning and polynomial chaos expansion effectively reduce the difficulty of high-dimensional uncertainty quantification problems and improve modeling efficiency. These improvements make the modeling method of the present invention more efficient and convenient, and reduce the requirements for the professional skills of modelers.

[0046] (4) Breaking through the limitations of uncertainty behavior prediction: The acquisition of uncertain characteristics in existing digital twin technologies often relies on a large amount of sample data for training and verification, resulting in high computational costs and insufficient generalization capabilities. The present invention combines manifold learning and polynomial chaos expansion methods to transform the uncertainty quantification problem of high-dimensional structural response fields into uncertainty analysis of low-dimensional features, thereby significantly reducing the demand for sample data. At the same time, this method estimates the uncertainty of low-dimensional popular features by solving the coefficients of the chaotic polynomial model, and then establishes an uncertainty mapping model from model parameters to popular features. This mapping model not only has high accuracy and robustness, but also can quickly and accurately realize the probabilistic behavior prediction of full-field performance, breaking through the limitations of existing technologies.

[0047] (5) Improving the accuracy of full-field performance probabilistic simulation: The probabilistic simulation prediction of the full-field performance response of turbine blades needs to comprehensively consider the interaction of multiple physical fields and multiple physical quantities. Existing models often have insufficient accuracy when dealing with such complex problems. The present invention realizes the rapid simulation prediction of the full-field uncertainty response by constructing a random moment mapping model of low-dimensional popular features and the full-field performance response of blade thermodynamics. The model quantifies the fluctuation and variation characteristics of the full-field response through sampling statistics such as mean, variance, skewness and kurtosis, and visually characterizes the uncertainty state of the blade's physical performance through the maximum entropy principle. This prediction method not only improves the accuracy and reliability of the prediction, but also enables designers to intuitively understand the performance of the blade under different working conditions and its uncertainty range, providing a more scientific basis for optimal design. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is the blade structure model in the present invention;

[0049] Figure 2 It is a schematic diagram of the comparison of the prediction accuracy of low-dimensional epidemic features in the present invention;

[0050] Figure 3 It is a schematic diagram of the random moment distribution field of the digital twin model in the present invention;

[0051] Figure 4 It is a visual schematic diagram of the probability distribution characteristics of the digital twin model in the present invention. DETAILED DESCRIPTION

[0052] The preferred invention of the present invention is described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used to explain the principles of the present invention together with the invention of the present invention.

[0053] like Figure 1-4As shown, the method of the present invention can be adapted to the construction of a digital twin model for an engine turbine blade, and can also be applied to any other structural model. In order to explain the technical solution of the present invention in more detail, the present invention takes a turbine blade as an example. Specifically, a data-driven digital twin modeling method for an engine turbine blade includes the following steps:

[0054] Step 1: Establish a thermal-mechanical coupling finite element model of the turbine blade, and determine the deviation model of the blade structural parameters based on engineering experience or experimental data. The model parameters include elastic modulus, Poisson's ratio, pressure side load, suction side load, thermal expansion coefficient, thermal conductivity, internal cooling convection coefficient, suction side convection coefficient, internal cooling ambient temperature and external blade ambient temperature.

[0055] The present invention provides a structural model of the engine turbine blade, such as Figure 1 There is a cooling duct inside the blade shown, which keeps the temperature of the blade within the limit of its material through cold air, and the pressure generated by the high-pressure gas is applied to the suction side and the pressure side respectively. The corresponding blade structure finite element model includes 11,794 units and 21252 nodes, and its high-dimensional response is 21252 dimensions. In addition, 12 arbitrary positions are given in the figure to visualize the response probability characteristics. The model parameters of the digital twin model constructed in this implementation mainly include elastic modulus, Poisson's ratio, pressure side load, suction side load, thermal expansion coefficient, thermal conductivity, internal cooling convection coefficient, suction side convection coefficient, internal cooling ambient temperature, external blade ambient temperature and other key parameters that need to be considered in blade design, such as environment, structure, and material. Table 1 gives the deviation values ​​of the above 10-dimensional model parameters and the corresponding distribution model.

[0056] Step 2: Based on the deviation model and finite element analysis, an optimized sampling strategy is used to obtain the optimal distribution sample points of the model parameters in the [0,1] cumulative probability space, as well as the corresponding full-field performance response data.

[0057] In the present invention, 100 samples are extracted according to the 10-dimensional variable distribution characteristics. First, according to the following minimization criterion, the sample data z with the best distribution in the [0,1] cumulative probability space is obtained. (i) ,i=1,2,…,100

[0058]

[0059] where d j is an ascending set of distances between any two points (d1, d2, ..., d j ,…,d η ) in the jth distance, J j This means that the distance between all pairs of points is exactly equal to d jThe optimal sample data in the parameter space can be obtained by the following formula:

[0060] x (i) =F -1 (z (i) ) (16)

[0061] Where F -1 is the optimization function based on the probability distribution model of parameter x in Table 1. Based on the above sampling strategy, the sample data corresponding to the model parameters of interest can be obtained Then, finite element simulations are performed to obtain the corresponding full-field performance response data Y = [y1, y2, ..., y 100 ],y i =[y1,y2,…,y 21252 ] T .

[0062] Step 3: Use machine learning methods to perform popular dimensionality reduction on high-dimensional full-field response data, extract low-dimensional features and corresponding popular functions according to the total empirical variance criterion, and then construct a popular structure from model parameters to full-field performance response.

[0063] In the present invention, according to the response data Y=[y1,y2,…,y 100 ] Solve for the corresponding mean vector and the covariance matrix Then, the corresponding popular feature parameters and feature vectors can be obtained by solving the following formula:

[0064]

[0065] The eigenvalues ​​after sorting by size are λ1≥λ2≥…≥λ m , the eigenvector is V = [v1,v2,…,v m ] According to the total empirical variance criterion Six main low-dimensional features and corresponding feature vectors can be determined, and their popular structure can be expressed as

[0066]

[0067] Based on the above popular structure, the 21252-dimensional data in the original space can be compressed into 6-dimensional data in the manifold space. The 6-dimensional popular features λ1≥λ2≥…≥λ6 and the feature vector V=[v1,v2,…,v6] can retain 99% of the information in the original data set.

[0068] Step 4: Perform uncertainty propagation analysis on the features after dimensionality reduction through the chaotic polynomial expansion method, solve the chaotic polynomial model coefficients to estimate the uncertainty of low-dimensional epidemic features, and establish an uncertainty mapping model from model parameters to epidemic features;

[0069] In the present invention, the uncertainty mapping model between 10-dimensional model parameters and 6-dimensional epidemic features is constructed by using the chaotic polynomial expansion method. First, a chaotic polynomial model between the model parameters and each feature is constructed. The core of the model is to determine the chaotic polynomial model coefficients according to the variables and response data through the following formula:

[0070]

[0071] In the formula, Represents the i-th high-dimensional response y i =[y1,y2,…,y 21252 ] T The low-dimensional characteristic response after compression by the popular structure. Due to the orthogonal characteristics of the chaotic polynomial, its origin moment can be directly obtained through the chaotic polynomial model coefficients

[0072]

[0073] The above process can only obtain the first-order and second-order origin moments. For the third-order and fourth-order origin moments and Quadratic and cubic values ​​of the available responses and The same process is used to construct the corresponding chaotic polynomial model and To verify the effectiveness of constructing a chaotic polynomial model of epidemic characteristics, the Monte Carlo method was used to extract 10,000 samples of 10-dimensional variables and then solve the corresponding epidemic characteristics and compare them with the model characteristics calculated by the chaotic polynomial model. Figure 2 The comparison results show that the constructed chaotic polynomial model has good prediction accuracy for the fluid characteristics of the full-field stress response of this blade.

[0074] Step 5: Based on the popular structure, a random moment mapping model of low-dimensional popular features and the full-field thermodynamic performance response of the blade is established to construct a digital twin model of the engine turbine blade, and to achieve rapid prediction of probabilistic characteristics such as fluctuations and variations in the full-field physical performance of the blade under multi-source deviations of model parameters.

[0075] In the present invention, the origin moment of the high-dimensional full-field response is obtained according to the origin moment of the low-dimensional popular feature, and a digital twin model is established to realize the probabilistic prediction of multi-field responses under multiple physical quantities. According to the following formula, the fourth-order origin moment of the full-field 21252-dimensional response can be analytically obtained through the first four-order origin moments of the 6-dimensional low-dimensional feature:

[0076]

[0077] In the formula is the expansion coefficient of the quadratic term, k = 1, 2, 3, 4. Based on this, a digital twin model of the engine turbine blade is established, which can realize the probabilistic behavior simulation from 10-dimensional model parameters to 21252-dimensional full-field response.

[0078] Step 6: Sample mean, variance, skewness, and kurtosis to quantify the fluctuation and variation characteristics of the full-field response and visually characterize the uncertainty state of the blade physical performance.

[0079] In the present invention, the origin moment of the response is converted into the corresponding random moments such as mean, standard deviation, skewness and kurtosis by the following formula, thereby realizing the visualization of the probability behavior of the response field.

[0080]

[0081] Figure 3 The mean field and standard deviation field of the blade response are given in . Based on the above random moments, the probability density function of the response can be solved by the maximum entropy principle, and its uncertain distribution characteristics can be visualized. In order to verify the effectiveness of building a digital twin model, the 10,000 samples obtained by the above Monte Carlo method are used to calculate the random moments of the response field. In addition, the random moments at 12 nodes are given in Table 2. It can be found that the digital twin model constructed by this patent can accurately predict the mean and standard deviation of all 12 positions, but there are errors in the skewness and kurtosis of some positions. An important factor leading to the above phenomenon is that some values ​​from the Monte Carlo method are relatively close to 0, such as the skewness at 9 positions, resulting in a large relative error. Figure 4 The probability distribution characteristics of the stress response at each node are visualized in . Figure 3 , Figure 4 From the comparison results, it can be seen that the digital twin model has a strong consistency with the Monte Carlo method results, which shows the effectiveness of constructing probabilistic behavior simulation of the digital twin model.

[0082] The method of the present invention establishes a non-invasive modeling framework by combining manifold learning and polynomial chaos expansion, and realizes the simulation of multiple physical quantities, multiple physical fields, and multiple probabilities of the structure by constructing a digital twin model of the model parameters and the full-field structural response of the blade. In addition, the method is based only on limited sample data without modifying the existing structural system, thereby greatly reducing the analysis and calculation cost of the complex mechanical behavior laws of the structure, providing an effective way to quickly and accurately predict the mechanical probabilistic behavior laws of the blade structure under arbitrary input and its deviation fluctuations, and also laying an important theoretical foundation for the refined design of turbine blades.

[0083] Embodiment 1: A data-driven digital twin modeling method for engine turbine blades

[0084] This embodiment uses a data-driven approach to build a digital twin model for a certain type of aircraft engine turbine blade, aiming to improve the accuracy and reliability of blade performance prediction. Turbine blades face complex mechanical and thermodynamic challenges in a high-temperature, high-pressure, and high-speed rotating working environment, so a modeling method that can fully consider the interaction of multiple physical fields and multiple physical quantities is needed.

[0085] Step 1: Establish a thermal-mechanical coupled finite element model of the turbine blade

[0086] First, based on the geometric dimensions and material properties of the turbine blades, a three-dimensional thermal-mechanical coupling model was established in the finite element analysis software. The model took into account key parameters such as the blade's elastic modulus, Poisson's ratio, and thermal expansion coefficient, and set corresponding boundary conditions such as pressure side load, suction side load, and internal cooling ambient temperature.

[0087] Step 2: Determine the blade model parameter deviation model

[0088] Based on engineering experience or experimental data, a deviation model is established for the key parameters in the model. For example, the elastic modulus and thermal expansion coefficient may have a certain fluctuation range due to differences in material batches, and these fluctuation ranges can be characterized by probability distribution models (such as normal distribution, lognormal distribution, etc.).

[0089] Step 3: Optimize sampling strategy to obtain sample data

[0090] An optimized sampling strategy (such as Latin hypercube sampling, maximizing minimum distance sampling, etc.) is used to generate a set of representative initial sample points in the parameter space. Then, these sample points are simulated using finite element analysis software to obtain the corresponding full-field performance response data, including temperature field, stress field, etc.

[0091] Step 4: Machine learning method for popular dimensionality reduction

[0092] The collected high-dimensional full-field response data is input into the machine learning algorithm (such as principal component analysis PCA, local linear embedding LLE, etc.) for popular dimensionality reduction. According to the total empirical variance criterion, low-dimensional features and corresponding popular functions are extracted to establish the popular structure of model parameters to full-field performance response.

[0093] Step 5: Chaotic polynomial expansion for uncertainty propagation analysis

[0094] The uncertainty propagation analysis of the features after dimensionality reduction is performed using chaotic polynomial expansion methods (such as generalized polynomial chaos GPC, sparse polynomial chaos SPC, etc.). The uncertainty of low-dimensional epidemic features is estimated by solving the chaotic polynomial model coefficients, and an uncertainty mapping model from model parameters to epidemic features is established.

[0095] Step 6: Establish a random moment mapping model to build a digital twin model

[0096] Based on the epidemic structure, a random moment mapping model of low-dimensional epidemic characteristics and blade thermodynamic full-field performance response is established. Through this model, the fluctuation and variation characteristics of the full-field physical performance of the blade can be quickly predicted under multi-source deviations of model parameters.

[0097] Step 7: Quantify and visualize the uncertainty of the full-field response

[0098] The sampling mean, variance, skewness and kurtosis are used to quantify the fluctuation and variation characteristics of the full-field response. The maximum entropy principle is used to solve the probability density function of the response, and the uncertainty state of the blade physical performance is visualized in the form of charts, cloud maps, etc.

[0099] This example successfully built a digital twin model of a certain type of aircraft engine turbine blade, and realized the performance prediction of the blade in a complex working environment. By optimizing the sampling strategy and applying machine learning methods, the computational cost and time cost in the modeling process were significantly reduced. At the same time, the establishment of the random moment mapping model improved the accuracy and reliability of the full-field performance response prediction, providing a scientific basis for the optimal design of the blade.

[0100] Example 2: Application example of a data-driven digital twin modeling method for engine turbine blades

[0101] This example applies the data-driven digital twin modeling method of engine turbine blades to actual engineering, aiming to improve the overall performance and reliability of the engine. Step 1 to Step 3: Model establishment and sample data collection

[0102] According to step 1 to step 3 in embodiment 1, a thermal-mechanical coupling finite element model of a turbine blade is established, and a deviation model of model parameters is determined. Then, an optimized sampling strategy is used to generate initial sample points in the parameter space, and a finite element simulation is performed to obtain full-field performance response data.

[0103] Step 4 to Step 6: Data processing and model building

[0104] In step 4, the local linear embedding (LLE) algorithm is used to perform popular dimension reduction on the high-dimensional full-field response data to extract key low-dimensional features and their popular functions. In step 5, the generalized polynomial chaos (GPC) method is used to perform uncertainty propagation analysis on the reduced-dimensional features, and an uncertainty mapping model from model parameters to popular features is established. Finally, in step 6, based on the popular structure and uncertainty mapping model, a random moment mapping model of low-dimensional popular features and blade thermodynamic full-field performance response is constructed to form a complete digital twin model.

[0105] Step 7: Performance prediction and optimized design

[0106] The constructed digital twin model is used to predict the performance of turbine blades under different working conditions. By adjusting the model parameters (such as pressure side load, suction side load, etc.), the changing trend of the blade's full-field performance response is observed, and the impact of different design schemes on blade performance is evaluated. Based on the prediction results, the blade design scheme is optimized and adjusted to improve its performance and reliability in actual work.

[0107] This embodiment successfully achieved accurate prediction and optimized design of the performance of a certain aircraft engine turbine blade by applying a data-driven digital twin model modeling method. The digital twin model not only improves the accuracy and reliability of performance prediction, but also provides a scientific basis and intuitive display for optimized design. Through the implementation of the optimized design scheme, the overall performance and reliability of the engine are effectively improved, and the R&D cost and time cycle are reduced, which has important engineering value and economic significance.

Claims

1. A data-driven digital twin modeling method for engine turbine blades. Features: The following steps are involved: Step 1: Establish a thermal-mechanical coupling finite element model of the turbine blade, and determine the deviation model of the blade structural parameters based on engineering experience or experimental data, wherein the model parameters include elastic modulus, Poisson's ratio, pressure side load, suction side load, thermal expansion coefficient, thermal conductivity, internal cooling convection coefficient, suction side convection coefficient, internal cooling ambient temperature, and external blade ambient temperature; Step 2: Based on the deviation model and finite element analysis, an optimized sampling strategy is used to obtain the optimal distribution sample points of the model parameters in the [0,1] cumulative probability space, as well as the corresponding full-field performance response data; Step 3: Use machine learning methods to perform popular dimension reduction on high-dimensional full-field response data, extract low-dimensional features and corresponding popular functions according to the total empirical variance criterion, and then construct a popular structure from model parameters to full-field performance response; Step 4: Perform uncertainty propagation analysis on the features after dimensionality reduction through the chaotic polynomial expansion method, solve the chaotic polynomial model coefficients to estimate the uncertainty of low-dimensional epidemic features, and establish an uncertainty mapping model from model parameters to epidemic features; Step 5: Based on the epidemic structure, a random moment mapping model of low-dimensional epidemic characteristics and the full-field performance response of blade thermodynamics is established to achieve rapid prediction of the probability characteristics of blade multi-physical field performance under multi-source deviations of model parameters; Step 6: The sampling mean, variance, skewness and kurtosis are used to quantify the probability characteristics of the full-field response, and the maximum entropy principle is used to visualize the probability states such as deviation or fluctuation of the blade physical performance.

2. According to claim 1, a data-driven digital twin modeling method for engine turbine blades, Features: The deviation model in step 1 uses a probability distribution model to characterize the uncertainty of each model parameter.

3. According to claim 1, a data-driven digital twin modeling method for engine turbine blades, Features: The step 2 first obtains the sample data z with the best distribution in the [0,1] cumulative probability space according to the following minimization criterion: (i) ,i=1,2,…,s where d j For any two points z (i) and z (j) The ascending set of distances (d 1 ,d 2 ,…,d j ,…,d η ) in the jth distance, J j Indicates that the distance between all pairs of sample points is exactly equal to d j The optimal sample data in the parameter space can be obtained by the following formula: x (i) =F -1 (z (i) ) (2) Where F -1 is a probability distribution optimization function based on parameter x. Based on the above sampling strategy, the model parameter x=[x 1 ,x 2 ,…x n ] Corresponding sample data Then, finite element simulation is performed to obtain the corresponding full-field performance response data Y = [y 1 ,y 2 ,…,y s ],y i =[y 1 ,y 2 ,…,y m ] T .

4. The data-driven digital twin modeling method for engine turbine blades according to claim 1, Features: The step 3 is performed as follows: In the formula By solving equation (3), the corresponding popular features and feature vectors can be solved. The eigenvalue after sorting is λ 1 ≥λ 2 ≥…≥λ m , the corresponding eigenvector is V = [v 1 ,v 2 ,…,v m ]. According to the total empirical variance criterion, The number of effective low-dimensional features can be determined. For the retained m'-dimensional low-dimensional features, its popular structure can be expressed as Based on the above popular structure, the large-dimensional data in the original space can be compressed into small-dimensional data in the manifold space, retaining most of the information in the large data set. At the same time, the high-dimensional response can also be reconstructed by superimposing the information contained in the m'-dimensional low-dimensional features through this structure.

5. The data-driven digital twin modeling method for engine turbine blades according to claim 1, Features: In step 4, the first second-order origin moment of the popular feature is obtained by the following formula Where β α (α=0,1,2,…,P-1) is the coefficient of the chaotic polynomial model of the t-th low-dimensional feature, which can be obtained by minimizing the following L 2 The norm residual is obtained In the formula, Represents the i-th high-dimensional response y i =[y 1 ,y 2 ,…,y m ] T The low-dimensional feature response after compression by the popular structure. and According to formula (6), the response and To construct the corresponding chaotic polynomial and get.

6. The data-driven digital twin modeling method for engine turbine blades according to claim 1, Features: The step 5 is to establish a random moment mapping model of low-dimensional popular features and full-field performance response according to the following formula: In the formula is the quadratic expansion coefficient, k = 1, 2, 3, 4. Based on this, a random moment mapping model of low-dimensional popular features and full-field performance responses is established. The first four-order origin moments of the high-dimensional responses of each dimension can be obtained through the first four-order origin moments of the low-dimensional feature responses.

7. The data-driven digital twin modeling method for engine turbine blades according to claim 1, Features: The step 1 is to visualize the uncertainty fluctuation and variation characteristics of the full-field response as follows: Based on the above random moments, the maximum entropy principle can be used to solve the probability density function of the response and realize the visualization of its probability distribution characteristics.

8. A data-driven engine turbine blade digital twin modeling method according to any one of claims 1 to 7, Features: The method further includes an adaptive adjustment step according to the complexity of the application scenario. By real-time monitoring of changes in the working environment parameters of the turbine blades, the boundary conditions and material property parameters in the thermal-mechanical coupling finite element model are dynamically adjusted to achieve accurate simulation of the probabilistic performance of the turbine blades under different working conditions.

9. A data-driven engine turbine blade digital twin modeling method according to any one of claims 1 to 7, Features: This method introduces automated process control in the modeling steps, automatically executing steps 1 to 6 through preset scripts or software tools, reducing human intervention, improving modeling efficiency and consistency, and ensuring the accuracy and repeatability of modeling details.

10. The data-driven digital twin modeling method for engine turbine blades according to claim 6, Features: The random moment mapping model iteratively optimizes the mapping relationship between low-dimensional popular features and full-field performance responses by introducing high-order chaotic polynomial expansion and reinforcement learning algorithm, thereby improving the accuracy and generalization ability of the random moment mapping model and achieving high-precision simulation prediction of the probabilistic behavior of full-field performance responses.

Citation Information

Patent Citations

  • An aero-engine turbine blade reliability digital twinning modeling method

    CN112507452A

  • Digital twin modeling method and prediction system for multi-fidelity data

    CN113591383A

  • Hybrid uncertainty mechanical property analysis method, electronic equipment and medium

    CN114925480A

  • Fan blade dynamic impact detection method and system based on digital twin modeling

    CN115859837A

  • Shield tunneling digital twin stratum construction method and system fusing multi-source data

    CN116227309A

Cited By

  • Real-time actual monitoring method and system for blisk of gas compressor

    CN120745340A

  • A method and system for real-time and effective monitoring of integral bladed disks of compressors

    CN120745340B

  • Engine performance simulation method and integrated simulation platform

    CN122021464A