Method and device for predicting dynamic reliability of high-temperature rotating components based on digital-analog drive

Abstract

This application discloses a method and apparatus for predicting the dynamic reliability of high-temperature rotating components based on numerical simulation, relating to the field of structural reliability analysis technology. The method includes obtaining the random variables and target cycle number of the high-temperature rotating component to be predicted; constructing a damage-threshold interference model based on interference theory where both damage accumulation and damage threshold follow a normal distribution; inputting the variables and cycle number into a trained data proxy model to obtain predicted damage parameters and their standard deviations; and calculating the predicted reliability based on the predicted values ​​using the damage-threshold interference model. The data proxy model is guided by a feedback function constructed based on the predicted values ​​and standard deviations, enabling iterative training and intelligent sample expansion. By integrating physical interference mechanisms with an actively learning-driven proxy model, this application achieves efficient prediction with minimal physical simulation samples while ensuring prediction accuracy, solving the problem of balancing computational efficiency and accuracy in the reliability assessment of high-temperature rotating components.

Description

Technical Field

[0001] This application relates to the field of structural reliability analysis technology, and in particular to a method and device for predicting the dynamic reliability of high-temperature rotating components based on numerical simulation. It further integrates multidisciplinary technologies such as data-driven modeling, physical mechanism simulation, closed-loop feedback control, and probabilistic statistical analysis. Its core applications are in key industrial fields such as aerospace, energy and power, nuclear power engineering, and metallurgy and chemical engineering, for the reliability design and evaluation of core structural components, such as aero-engine turbine disks and gas turbine blades, which withstand complex service environments including high temperatures, cyclic loads, and corrosion. Background Technology

[0002] In strategic emerging industries such as aerospace, energy and power, and nuclear power, the service safety of high-temperature rotating components is crucial for ensuring the long-term reliable operation of critical equipment. These components operate under extreme and complex environments such as high temperatures and cyclic loads for extended periods, and their failure can lead to serious consequences. Therefore, accurately predicting their dynamic reliability at different cycle counts is a key technical requirement in engineering design and operation and maintenance.

[0003] Purely physics-driven methods (such as Monte Carlo simulations) rely on physical models for computational accuracy but suffer from extremely low computational efficiency, making them unsuitable for rapid iteration and real-time evaluation in engineering. While purely data-driven methods (based on machine learning surrogate models) offer high computational efficiency, they are hampered by the high cost and difficulty of obtaining high-quality training samples in high-temperature rotating component scenarios. This results in poor model generalization ability and insufficient prediction accuracy with small sample sizes, as well as poor extrapolation due to a lack of physical mechanism constraints. Although coupled physics and data-driven methods have been proposed, existing invasive modeling and other approaches are difficult to apply to high-temperature rotating components with extremely complex physical simulation processes (such as those involving thermo-mechanical coupling analysis).

[0004] Therefore, there is an urgent need for a dynamic reliability prediction method for high-temperature rotating components based on numerical modeling to solve the problem of the inability to balance computational efficiency and accuracy. This method should significantly reduce the reliance on high-cost physical simulations or experiments while ensuring prediction accuracy, and overcome the technical problems of insufficient model generalization and extrapolation capabilities under small sample conditions. This would improve the real-time performance and engineering applicability of dynamic reliability assessment for high-temperature rotating components. Summary of the Invention

[0005] The purpose of this application is to provide a method and device for predicting the dynamic reliability of high-temperature rotating components based on digital model driving, which can achieve the best balance between computational accuracy and computational efficiency in reliability calculation.

[0006] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for predicting the dynamic reliability of high-temperature rotating components based on digital model-driven methods, including: Obtain the random variables and target cycle number of the high-temperature rotating component to be predicted; the random variables include material property parameters, geometric dimension parameters, and load parameters; A damage-threshold interference model is constructed based on the interference theory that both damage accumulation and damage threshold follow a normal distribution. The random variables of the high-temperature rotating component to be predicted and the target number of cycles are input into a trained data proxy model to obtain the predicted damage parameters and the standard deviation of the predicted damage parameters of the high-temperature rotating component at the target number of cycles. The data proxy model is obtained by iteratively training a preset machine learning proxy model based on an initial training dataset until a preset termination condition is met. The initial training dataset includes multiple sets of triplet data of random variable-cycle number-damage parameter baseline value. The preset termination condition includes the value of the feedback function being less than a preset threshold. The feedback function is constructed based on the predicted damage parameters and the standard deviation of the predicted damage parameters. Based on the predicted damage parameters, the predicted reliability of the high-temperature rotating component under the target cycle number is obtained through the damage-threshold interference model; the predicted reliability is the probability that the accumulated damage is less than the damage threshold.

[0007] Secondly, this application provides a dynamic reliability prediction device for high-temperature rotating components based on digital model driving, comprising: The data acquisition module is used to acquire the random variables and target cycle number of the high-temperature rotating component to be predicted; the random variables include material property parameters, geometric dimension parameters, and load parameters. The model building module is used to construct a damage-threshold interference model based on the interference theory that both damage accumulation and damage threshold follow a normal distribution. The damage prediction module is used to input the random variables of the high-temperature rotating component to be predicted and the target cycle number into a trained data proxy model to obtain the predicted damage parameters and the standard deviation of the predicted damage parameters of the high-temperature rotating component at the target cycle number. The data proxy model is obtained by iteratively training a preset machine learning proxy model based on an initial training dataset until a preset termination condition is met. The initial training dataset includes multiple sets of random variable-cycle number-damage parameter baseline triplet data. The preset termination condition includes the value of the feedback function being less than a preset threshold. The feedback function is constructed based on the predicted damage parameters and the standard deviation of the predicted damage parameters. The reliability acquisition module is used to obtain the predicted reliability of the high-temperature rotating component under the target cycle number based on the predicted damage parameter value and through the damage-threshold interference model; the predicted reliability is the probability that the accumulated damage is less than the damage threshold.

[0008] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method and apparatus for predicting the dynamic reliability of high-temperature rotating components based on numerical modeling. By acquiring the random variables and target cycle number of the high-temperature rotating component to be predicted, it solves the problem of the difficulty in uniformly quantifying and modeling multi-source uncertainties, and realizes a comprehensive characterization of material property dispersion, geometric tolerance, and load fluctuation, providing accurate input for probabilistic analysis. By constructing a damage-threshold interference model based on the interference theory that both damage accumulation and damage threshold follow a normal distribution, it solves the problem of the lack of probabilistic expression of the dynamic competition relationship between the damage evolution process and the failure threshold of high-temperature rotating components under cyclic loading, and realizes the establishment of a physical framework for reliability calculation based on probabilistic interference theory. This application inputs the random variables of the high-temperature rotating component to be predicted and the target cycle number into a trained data proxy model to obtain predicted damage parameters and their standard deviations. This model is obtained through iterative training guided by a feedback function, overcoming the dual technical bottlenecks of low computational efficiency in traditional Monte Carlo methods and insufficient prediction accuracy of purely data-driven methods under small sample conditions. It significantly reduces reliance on high-cost physical simulations or experiments while maintaining prediction accuracy. Furthermore, by calculating predicted reliability based on the predicted damage parameters using a damage-threshold interference model, it solves the problem of difficulty in quickly and accurately assessing the dynamic reliability of high-temperature rotating components, achieving efficient and accurate output of service reliability indicators from multi-source random inputs. This application effectively balances computational accuracy and efficiency, significantly improving the engineering applicability and timeliness of reliability assessment for high-temperature rotating components. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a method for predicting the dynamic reliability of high-temperature rotating components based on digital model driving, provided in one embodiment of this application.

[0011] Figure 2 This is a flowchart illustrating a method for predicting the dynamic reliability of high-temperature rotating components based on digital model driving, provided in another embodiment of this application.

[0012] Figure 3 A schematic diagram of the probabilistic lifetime prediction model structure is provided for one embodiment of this application.

[0013] Figure 4 A comparison chart of the average reliability calculation error results provided in an embodiment of this application.

[0014] Figure 5 A two-dimensional finite element model and equivalent stress result cloud diagram of a tenon joint structure provided in an embodiment of this application; wherein, Figure 5 (a) is a schematic diagram of a two-dimensional finite element model of a tenon joint structure; Figure 5 (b) in the diagram is the equivalent stress result contour plot.

[0015] Figure 6 This is a schematic diagram illustrating the iterative process of a high-temperature rotating component dynamic reliability prediction method based on digital model driving, provided in an embodiment of this application; wherein, Figure 6 (a) in the diagram illustrates the change of the iteration quantity with the number of iterations; Figure 6 (b) in the figure represents the damage parameter G. P A schematic diagram illustrating how the statistics change with the number of iterations.

[0016] Figure 7 A graph showing the variation in the minimum sample size required to determine the reliability of a tenon joint structure under different cycle counts provided in an embodiment of this application using a physical drive method.

[0017] Figure 8 A comparison diagram of the dynamic reliability of the tenon joint structure provided in one embodiment of this application.

[0018] Figure 9 This is a functional module diagram of a high-temperature rotating component dynamic reliability prediction device based on digital model driving, provided in an embodiment of this application. Detailed Implementation

[0019] First, some technical terms involved in the embodiments of this application will be introduced.

[0020] Calculating the dynamic reliability of high-temperature rotating components faces multiple technical challenges. On the one hand, the failure mechanisms of structures under high-temperature environments exhibit significant complexity and coupling, typically involving the synergistic effects of multiple mechanisms such as fatigue damage accumulation, creep deformation, and fatigue-creep coupled failure, with the damage evolution process dynamically changing with the number of cycles. On the other hand, the reliability of components is comprehensively affected by multiple sources of random variables, including the dispersion of material mechanical properties (such as statistical fluctuations in the elastic modulus, fatigue strength coefficient, and creep constitutive parameters of high-temperature alloys), machining tolerances of geometric dimensions, and random fluctuations in service loads (such as engine speed fluctuations and load peak variations), resulting in strong randomness and uncertainty in the failure of high-temperature rotating components. Therefore, the core objective of dynamic reliability calculation for high-temperature rotating components is to achieve a balance between computational accuracy and computational efficiency while considering multi-source randomness and dynamic damage evolution.

[0021] Physics-driven methods establish damage evolution models based on well-defined physical mechanisms. They utilize large-scale physical simulations (Monte Carlo Simulation (MCS)) to traverse the input space of multi-source random variables, thereby calculating the reliability of components at different cycle counts. Their advantage lies in the clear physical meaning, and the computational accuracy depends on the rationality of the physical model and the detail of the simulation. However, a fatal flaw of this type of method is its extremely low computational efficiency: obtaining the damage parameters corresponding to each sample requires complex high-temperature mechanical simulations. For Monte Carlo simulations requiring tens or even hundreds of thousands of samplings, the computation cycle often lasts for weeks or even months, making it difficult to meet the rapid iteration requirements of engineering design, and even more unsuitable for real-time safety assessments of in-service structures (high-temperature rotating components).

[0022] To address the efficiency bottleneck of purely physical-driven methods, researchers have proposed using machine learning to construct surrogate models. These models are trained with a small number of physical simulation samples, establishing an implicit mapping between multi-source random variables and damage parameters, thereby enabling rapid prediction of the reliability of high-temperature rotating components. The core advantage of this approach is its high computational efficiency; once the model is trained, reliability prediction can be completed in milliseconds. However, in the scenario of high-temperature rotating components, purely data-driven methods have fatal flaws: first, sample acquisition is difficult; structural experiments and physical simulations under high-temperature environments are extremely costly, making it difficult to obtain a sufficient number of high-quality training samples, resulting in poor model generalization ability and insufficient prediction accuracy; second, there is a lack of physical mechanism constraints, meaning the model can only learn statistical regularities from the training samples. Under cycle counts or load conditions exceeding the training sample range, the prediction results show significant deviations and poor extrapolation.

[0023] Addressing the respective advantages and disadvantages of physics-driven and data-driven approaches, the coupled-driven method combines the strengths of both to achieve a synergistic effect. A typical analysis method is invasive modeling, which incorporates the constraints of the physical model into the data-driven process. This allows the data-driven process to achieve high prediction accuracy even with small sample sizes, thus minimizing the number of physical simulation calls while maintaining high prediction accuracy. This method is well-suited for cases with extremely simple physical equations. However, the physical simulation of high-temperature rotating structures is extremely complex, involving thermo-mechanical coupling analysis, making it difficult to use this invasive analysis method to assess structural reliability. Developing a novel coupled-driven method for dynamic reliability assessment of high-temperature rotating components remains a current research challenge.

[0024] 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. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, this application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0026] In one exemplary embodiment, such as Figure 1 As shown, a dynamic reliability prediction method for high-temperature rotating components based on digital model driving is provided, including the following steps 201 to 204. Wherein: Step 201: Obtain the random variables and target cycle number of the high-temperature rotating component to be predicted; the random variables include material property parameters, geometric dimensional parameters, and load parameters. The cycle number is the total number of complete "load-unload-reload" cycles experienced by the high-temperature rotating component during its service life. It is a core time indicator for measuring the cumulative amount of fatigue / creep-fatigue damage, usually represented by a positive integer N. Each additional cycle corresponds to the incremental expansion of microcracks or creep strain inside the material until critical damage is reached and failure occurs.

[0027] Step 202: Construct a damage-threshold interference model based on the interference theory that both damage accumulation and damage threshold follow a normal distribution.

[0028] Step 203: Input the random variable of the high-temperature rotating component to be predicted and the target cycle number into the trained data proxy model to obtain the predicted damage parameter value and the standard deviation of the predicted damage parameter value of the high-temperature rotating component under the target cycle number; the data proxy model is obtained by iteratively training a preset machine learning proxy model based on an initial training dataset until a preset termination condition is met; the initial training dataset includes multiple sets of random variable-cycle number-damage parameter baseline value triplet data; the preset termination condition includes the value of the feedback function being less than a preset threshold; the feedback function is constructed based on the predicted damage parameter value and the standard deviation of the predicted damage parameter value.

[0029] Step 204: Based on the predicted damage parameter values, the predicted reliability of the high-temperature rotating component to be predicted under the target cycle number is obtained through the damage-threshold interference model; the predicted reliability is the probability that the accumulated damage is less than the damage threshold.

[0030] By implementing steps 201 to 204 above, this application can fully leverage the advantages of data-driven and physics-driven approaches, achieving a balance between computational efficiency and computational accuracy, and can well meet the needs of engineering applications.

[0031] In another exemplary embodiment of this application, the training process of the data proxy model specifically includes: Step S1: Based on the initial training dataset, use the random variable of the high-temperature rotating component and the number of cycles as input to the data proxy model, output the predicted damage parameter value and the standard deviation of the predicted damage parameter value corresponding to the high-temperature rotating component, train the data proxy model until the difference between the predicted damage parameter value and the corresponding benchmark damage parameter value is less than the first preset threshold, and the standard deviation of the predicted damage parameter value is less than the second preset threshold, then stop training and obtain the current data proxy model.

[0032] Step S2: Generate a preset number of candidate sample points using the Latin hypercube sampling method to obtain a candidate sample set; the candidate sample points include random variables and cycle numbers.

[0033] Step S3: Based on the candidate sample set, obtain the predicted damage parameter values ​​and the standard deviation of the predicted damage parameter values ​​for each candidate sample point in the candidate sample set through the current data proxy model.

[0034] Step S4: Calculate the feedback function value of each candidate sample point in the candidate sample set using the feedback function to obtain the candidate feedback function value set.

[0035] Step S5: If the largest feedback function value in the candidate feedback function value set is greater than the third preset threshold, then perform high-temperature rotating component physical simulation on the candidate sample point corresponding to the largest feedback function value to obtain the corresponding damage parameter baseline value; form a triplet data set with the candidate sample point corresponding to the largest feedback function value and the corresponding damage parameter baseline value, add it to the initial training dataset, and return to step S1. The established feedback function essentially determines whether adding candidate samples can maximize the improvement of the output difference of the current sample set. The greater the output sample difference, the richer the data-driven learning information, and the closer the prediction effect is to the actual physical simulation result. Therefore, only the candidate sample that makes the feedback function value the largest in each iteration is selected for physical simulation, and the simulation result is added to the sample set to further update the data-driven model.

[0036] Step S6: If the largest feedback function value in the candidate feedback function value set is less than or equal to the third preset threshold, then the current data proxy model is used as the trained data proxy model. The iteration termination condition is controlled by the value of the feedback function. As the iteration proceeds, the difference in the sample set will gradually decrease, that is, the value of the feedback function will gradually decrease. When the value is lower than the defined threshold, the iteration is considered complete, and the data-driven model is considered ready for subsequent structural reliability calculations.

[0037] In another exemplary embodiment of this application, based on the predicted damage parameter values, the predicted reliability of the high-temperature rotating component to be predicted at the target number of cycles is obtained through a damage-threshold interference model, specifically including: The prediction reliability of the high-temperature rotating component under the target cycle number is obtained using the following damage-threshold interference model: .

[0038] in, This indicates the predicted reliability of a high-temperature rotating component at cycle N. The cumulative distribution function representing the standard normal distribution; Indicates the predicted value of damage parameters The reliability index is the highest reliability index. Indicates the predicted value of damage parameters The probability density function.

[0039] In another exemplary embodiment of this application, the formula for calculating the reliability index is: .

[0040] in, A This represents the reciprocal of the slope of the linear function of the damage parameters of high-temperature rotating components obtained by fitting experimental data and the mean life. A 1 represents the reciprocal of the slope of the linear function of the damage parameters of high-temperature rotating components fitted based on experimental data and the lower limit of the one-sided standard deviation of life. A and A 1. Calibration is required based on the fatigue and creep-fatigue test results of multiple sets of parallel specimens under different working conditions.

[0041] In another exemplary embodiment of this application, the feedback function is: .

[0042] .

[0043] in, Represents candidate sample points The feedback function value; and represents the predicted damage parameter value and the standard deviation of the predicted damage parameter value for the candidate sample point, respectively; The probability density function representing the standard normal distribution; The cumulative distribution function representing the standard normal distribution; and These represent the upper and lower bounds of standardization, respectively. and These represent the maximum and minimum baseline damage parameter values ​​in the initial training dataset, respectively. The core of the coupling mechanism between physics-driven and data-driven approaches lies in constructing a feedback function to establish iterative collaboration between the two approaches, achieving the optimal balance between computational accuracy and efficiency. The feedback function guides the simulation direction of the physics-driven approach based on the data-driven results, selecting the samples that provide the greatest improvement to the data-driven approach from a large-scale input sample space for physics simulation. The physics simulation results further improve the prediction performance of the data-driven model. By establishing this feedback function, the collaborative invocation of data-driven and physics-driven approaches is realized.

[0044] In another exemplary embodiment of this application, the data proxy model includes Artificial Neural Network (ANN), Radial Basis Function (RBF), Extreme Learning Machine (ELM), Support Vector Regression (SVR), or Gaussian Process Regression (GPR), etc. The theoretical basis of the Gaussian Process Regression model is probability theory and Bayesian information updating. Compared with other data-driven proxy models, it can simultaneously output the predicted mean and predicted variance, providing more prediction information. Therefore, the Gaussian Process Regression model is recommended as the data-driven proxy model.

[0045] In another exemplary embodiment of this application, based on the predicted damage parameter values, the predicted reliability of the high-temperature rotating component to be predicted at the target cycle number is obtained through a damage-threshold interference model. This is followed by comparing the results with large-sample calculations driven by physical methods to verify the correctness of the coupled driving method, which can be expressed as: .

[0046] in, This indicates the sample size in the Monte Carlo simulation. The baseline value of the damage parameter for the j-th sample needs to be obtained through physical simulation. The sample size is determined using a statistical error with a 95% confidence level, and can be expressed as: .

[0047] in, express The standard deviation needs to be determined by sampling. This represents the maximum allowable error, which is 0.1 times the current estimated probability of failure. That is, if the current estimated reliability is 0.90, then the true value has a 95% chance of falling within the interval [0.89, 0.91]. This indicates the probability of failure.

[0048] The following example illustrates this application using a specific digital-analog-driven dynamic reliability prediction process for high-temperature rotating components.

[0049] This application establishes a physics-driven damage-threshold interference model to analyze the reliability of a structure at different cycle counts; it also establishes a data-driven surrogate model to construct an implicit mapping relationship between multi-source random variable inputs and damage variable outputs; it establishes a coupling mechanism between physics-driven and data-driven approaches, constructs a feedback function, and realizes collaborative iterative calls between data and physics; it determines the iteration termination condition and calculates the reliability of the high-temperature rotating component at different cycle counts based on the final iteration results; and it compares the results with large-sample calculations driven by physics to verify the correctness of the coupled-driven method. The dynamic reliability calculation method for high-temperature rotating components provided in this application can fully leverage the advantages of both data-driven and physics-driven approaches, achieving a balance between computational efficiency and accuracy, and can well meet the needs of engineering applications.

[0050] like Figure 2 As shown in the figure, this embodiment provides a dynamic reliability prediction method for high-temperature rotating components based on digital analog driving, which includes the following steps: (1) Establish a physical-driven damage-threshold interference model to analyze the reliability of the structure (high-temperature rotating component) under different cycles.

[0051] Under the same operating conditions, the failure lifetime of the structure exhibits significant dispersion. Therefore, P is first constructed. f -G p -N f Line, such as Figure 3 As shown, a schematic diagram of the probabilistic lifetime prediction model is provided. The solid line in the middle of the figure represents the linear function of mean lifetime and damage parameter, and the dashed line on the left represents the linear function of one-sided standard deviation lifetime lower limit and damage parameter. A and A 1 represents the parameters of the probabilistic lifetime prediction model. A It is the reciprocal of the slope of the middle solid line. A 1 represents the reciprocal of the slope of the dashed line on the left. The failure lifetime under given damage parameters is considered a random variable, assumed to follow a normal distribution, with the midpoint lifetime being... The lower limit of lifespan is one standard deviation. They are represented as follows: .

[0052] Therefore, given the damage parameters, the standard deviation of the failure life It can be represented as: .

[0053] According to the linear damage accumulation criterion, the damage accumulation under any number of cycles N It can be represented as: .

[0054] When the number of cycles is exactly equal to the failure lifetime, the cumulative damage at this point is considered to be the damage threshold. Therefore, the damage threshold mean and standard deviation They are respectively: .

[0055] .

[0056] Since a linear damage accumulation rule is used, the expected value (mean) of damage accumulation under any other cycle number can be obtained based on the mean and standard deviation of the damage threshold. and standard deviation respectively for: .

[0057] .

[0058] When the accumulated damage is less than the damage threshold, the system is considered reliable and will not fail. Since both the accumulated damage and the damage threshold are normally distributed, given G... p The following reliability calculation formula is: .

[0059] .

[0060] For high-temperature rotating components, it is necessary to consider individual differences. For example, 1000 engine disks from the same batch may have random variations in material mechanical properties, structural geometry, and service load conditions. These parameters will directly affect G. p Therefore, in the reliability analysis of actual high-temperature rotating components, it is necessary to consider G. p Treating it as a random variable, the reliability of the structure in a certain cycle N can be expressed as the following total probability: .

[0061] The above formula is essentially In G p The expectation over the distribution domain cannot be analytically expressed and can only be solved by sampling, using the sample mean as the expected value (predicted value).

[0062] (2) Establish a data-driven proxy model and construct an implicit mapping relationship between the input of multi-source random variables and the output of damage variables.

[0063] The reliability calculation of physically driven high-temperature rotating components requires an extremely large sample size, which can be expressed as: .

[0064] Get each G pj Both require physical simulation to obtain, therefore an approximate function needs to be constructed to replace the multi-source uncertainties in the input and G. p This physical simulation process is output, and data-driven methods have significant advantages in function approximation. This application recommends using the GPR model because, compared with other data-driven models, it can provide not only the predicted mean but also the predicted variance, providing users with more prediction information.

[0065] (3) Establish a coupling mechanism between physical drive and data drive, construct a feedback function, and realize the collaborative iterative call between data and physics.

[0066] This application constructs a feedback mechanism to selectively invoke physical simulations based on data-driven prediction results, rather than performing physical simulations aimlessly for every sample point. Because... It has a clear mathematical expression, therefore the key to the calculation lies in G. p Approximating a distribution. To approximate a distribution, the data-driven model needs to learn as many output sample features as possible. Therefore, this application constructs a feedback function based on improved sample dissimilarity: Existing datasets : . , and Representing datasets respectively The random variables and cycle number of the first, second, and nth sample points in the dataset; , and Representing datasets respectively The predicted damage parameters for the first, second, and nth sample points.

[0067] Current difference : . and Representing datasets respectively The maximum and minimum damage parameter benchmark values ​​are defined in the data.

[0068] New candidate sample point: A new candidate sample point in the input space. True output of unknown candidate sample points Damage parameters : Prediction requires a data-driven approach. This represents a data proxy model.

[0069] New dissimilarity: The dataset becomes The new degree of difference becomes: .

[0070] Difference Improvement: Only when the new difference Greater than the current difference Only then was it considered that the sample difference had improved. ,because Unknown therefore based on The range of values ​​will determine the degree of difference. Divided into the following three sections: .

[0071] Expected improvement in difference: due to Unknown, therefore For random variables, through The degree of improvement in the expected value quantification of the difference: .

[0072] .

[0073] .

[0074] .

[0075] Selected best candidate sample point x : .

[0076] in, F1 represents the feedback function; F1 represents the new candidate sample points. The predicted damage parameter values ​​exceed the expected improvement of the maximum damage parameter baseline value in the dataset; F2 represents the new candidate sample points. The expected improvement when the predicted damage parameter value is lower than the baseline value of the largest damage parameter in the dataset; F3 represents the new candidate sample point The expected improvement in the predicted damage parameters lies between the maximum and minimum baseline damage parameter values ​​in the dataset; This represents the candidate sample set. A predetermined number of candidate sample points are generated using the Latin hypercube sampling method to obtain the candidate sample set.

[0077] By constructing the above feedback mechanism, data-driven approaches can guide the direction of physics-driven simulations, and physics-driven results can optimize data-driven predictions, ultimately achieving a closed-loop feedback between the two. This allows for the maximum improvement of data-driven approaches with the lowest cost of physics simulations, balancing computational efficiency and accuracy.

[0078] To verify the effectiveness of this feedback mechanism, four numerical examples from different dimensions are used for illustration. It is a known mathematical test function, which was used as a "virtual physics simulator" in the verification experiment, and its return value is... It simulates the "damage parameter values" that can only be obtained through high-cost simulation in real engineering problems. , , and These represent two-dimensional test functions, three-dimensional test functions, five-dimensional test functions, and ten-dimensional test functions, respectively. .

[0079] .

[0080] .

[0081] .

[0082] The input variables of the four functions with different dimensions are independent and identically distributed standard normal random variables. The constructed feedback mechanism is used to approximate the probability distribution of the true function output. Finally, the distribution approximation results of the data-driven method and the method proposed in this application under the same sample size are listed in Tables 1-4.

[0083] Table 1 Comparison of Calculation Results of Two-Dimensional Test Functions

[0084] Table 2 Comparison of Calculation Results of Three-Dimensional Test Functions

[0085] Table 3 Comparison of Calculation Results of Five-Dimensional Test Function

[0086] Table 4 Comparison of Calculation Results of the Ten-Dimensional Test Function

[0087] The mean, standard deviation, kurtosis, and skewness are used to quantify how well each model approximates the true distribution. The final average relative error of each model is as follows: Figure 4 As shown, the calculation results of this application show that compared with the large-scale simulation driven by pure physics, the average relative error of this application is only about 5%, which is far better than other data-driven models.

[0088] (4) Determine the iteration termination condition and calculate the reliability of the high-temperature rotating component at different cycles based on the final iteration results.

[0089] Based on the analysis results of multiple numerical examples, when the numerical iteration of the feedback function reaches approximately 0.2, it can be considered that the sample difference can hardly be improved further, and the iteration can be terminated. A practical case study is used to illustrate this, where the load mainly consists of centrifugal force generated by blade rotation. This is simulated by applying a pressure load perpendicular to the tenon-groove contact surface. Figure 5 As shown in (a) above. Finite element simulation based on the applied constraints and load conditions can yield the stress field distribution of the structure, as shown below. Figure 5 As shown in (b), the color from red to blue represents stress decreasing, with units of MPa. Due to significant differences in the mechanical properties of materials from different batches, the material parameters in the constitutive model should be considered random variables. Simultaneously, the blade mass and rotational speed jointly determine the stress on the tenon, and considering individual differences, the blade mass and rotational speed also exhibit significant dispersion. Because the machining tolerance is extremely small, geometric variables such as the blade rotation radius, tenon contact area, and tenon angle are treated as constants. However, the tenon joint at critical locations may be significantly affected by geometry, primarily influencing its stress concentration factor; therefore, the stress concentration factor is considered a random variable. All the random variables mentioned above are assumed to follow a normal distribution, and their distribution parameters are shown in Table 5 below.

[0090] Table 5 Random variable input distribution parameters for mortise and tenon structure cases

[0091] The analysis using the data-physical coupling driven method of this application shows that the iteration quantity (expected improvement in sample variability) changes with the number of iterations as follows: Figure 6As shown in (a), the initial sample size is 30. Only one sample is added for each iteration in the physical simulation. The iteration stops when the expected improvement in sample difference is less than 0.15 after 60 iterations. During the iteration process, the damage parameter G... p The iterative approximation process of the distribution is as follows Figure 6 As shown in (b) in the figure, it can be seen that after 20 iterations, the statistical moments of the predicted distribution tend to be stable, proving that the iterative process converges.

[0092] After the iteration ends, the reliability of the structure under different cycles can be calculated using data-driven methods. That is, data-driven methods are used to replace the physical simulation process. By generating a large number of samples, the reliability value can be quickly calculated based on data-driven methods. The reliability calculation results of the tenon structure under different cycles in this case are shown in Table 6 below.

[0093] Table 6 Summary of calculation results of the single data-driven method and the coupled-driven method of this application

[0094] (5) Compare with the large sample calculation results of physical driving to verify the correctness of the coupling driving method.

[0095] To verify the accuracy of the calculation results from the pure data-driven method and the coupled-driven method of this application, large-sample calculations using physical-driven methods are required. The sample size is determined based on the statistical error of the 95% confidence interval. .

[0096] Therefore, the lower the failure probability, the larger the sample size required. In this case, the required physical simulation sample size and failure probability estimate for the tenon joint structure at different cycle counts are as follows: Figure 7 As shown, it can be seen that the required sample size exceeds 10 when N=500 cycles. 4 That is, more than 10 4 Only a partial physical simulation (MCS) can accurately estimate reliability, which obviously has an extremely high computational cost. Comparing the analysis results of data-driven and coupled-driven approaches with the large-sample calculation results of this physical-driven approach yields the following results: Figure 8 As shown, compared with other data-driven models such as ANN and SVR, the data-physics closed-loop feedback coupling method proposed in this application yields the best results in large-sample calculations with purely physics-driven methods. Furthermore, the sample size used in this application is the same as that of a single data-driven method, requiring only 90 physical simulation calculations. This demonstrates that the data-physics closed-loop feedback coupling method proposed in this application effectively leverages the advantages of both data-driven and physics-driven approaches, achieving a balance between computational efficiency and accuracy, and possesses strong engineering application value.

[0097] Based on the same inventive concept, this application also provides a device for predicting the dynamic reliability of high-temperature rotating components based on analog-driven methods, which implements the aforementioned method for predicting the dynamic reliability of high-temperature rotating components based on analog-driven methods. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the device for predicting the dynamic reliability of high-temperature rotating components based on analog-driven methods provided below can be found in the limitations of the analog-driven method for predicting the dynamic reliability of high-temperature rotating components described above, and will not be repeated here.

[0098] In one exemplary embodiment, such as Figure 9 As shown, a dynamic reliability prediction device for high-temperature rotating components based on digital model driving is provided, comprising: The data acquisition module 301 is used to acquire the random variables and target cycle number of the high-temperature rotating component to be predicted; the random variables include material property parameters, geometric dimension parameters and load parameters.

[0099] Model building module 302 is used to build a damage-threshold interference model based on interference theory, where both damage accumulation and damage threshold follow a normal distribution.

[0100] The damage prediction module 303 is used to input the random variables of the high-temperature rotating component to be predicted and the target cycle number into a trained data proxy model to obtain the predicted damage parameters and the standard deviation of the predicted damage parameters of the high-temperature rotating component at the target cycle number. The data proxy model is obtained by iteratively training a preset machine learning proxy model based on an initial training dataset until a preset termination condition is met. The initial training dataset includes multiple sets of random variable-cycle number-damage parameter baseline triplet data. The preset termination condition includes the value of the feedback function being less than a preset threshold. The feedback function is constructed based on the predicted damage parameters and the standard deviation of the predicted damage parameters.

[0101] The reliability acquisition module 304 is used to obtain the predicted reliability of the high-temperature rotating component under the target cycle number based on the predicted damage parameter value and through the damage-threshold interference model; the predicted reliability is the probability that the accumulated damage is less than the damage threshold.

[0102] In summary, this application does not require users to have extensive knowledge of physical simulation. By establishing a feedback mechanism between data-driven and physics-driven approaches, it achieves synergistic interaction between the two. Specifically, based on the output of the data-driven approach, the most valuable sample points are selectively chosen and used for physical simulation calculations. The results of the physical simulation calculations are then used to train the data-driven approach again, thus forming a closed-loop feedback. This analytical method can significantly reduce the cost of performing physical simulations at worthless sample points, thereby greatly improving computational efficiency while ensuring computational accuracy.

[0103] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores reliability prediction processing data for high-temperature rotating components. The I / O interfaces of the computer device are used for exchanging information between the processor and external devices. The communication interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a reliability prediction method for high-temperature rotating components.

[0104] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0105] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0106] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0107] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0108] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0109] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0110] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0111] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting the dynamic reliability of high-temperature rotating components based on digital model driving, characterized in that, include: Obtain the random variables and target cycle number of the high-temperature rotating component to be predicted; The random variables include material property parameters, geometric dimensional parameters, and load parameters; A damage-threshold interference model is constructed based on the interference theory that both damage accumulation and damage threshold follow a normal distribution. The random variable of the high-temperature rotating component to be predicted and the target number of cycles are input into the trained data proxy model to obtain the predicted damage parameters and the standard deviation of the predicted damage parameters of the high-temperature rotating component under the target number of cycles. The data proxy model is obtained by iteratively training a preset machine learning proxy model based on an initial training dataset until a preset termination condition is met. The initial training dataset includes multiple sets of triplet data of random variable-cycle number-damage parameter baseline value. The preset termination condition includes the value of the feedback function being less than a preset threshold. The feedback function is constructed based on the predicted values ​​of the damage parameters and the standard deviation of the predicted values ​​of the damage parameters; The training process of the data proxy model specifically includes: Step S1: Based on the initial training dataset, use the random variable of the high-temperature rotating component and the number of cycles as the input of the data proxy model, output the predicted damage parameter value and the standard deviation of the predicted damage parameter value corresponding to the high-temperature rotating component, train the data proxy model until the difference between the predicted damage parameter value and the corresponding benchmark damage parameter value is less than the first preset threshold, and the standard deviation of the predicted damage parameter value is less than the second preset threshold, then stop training and obtain the current data proxy model. Step S2: Generate a preset number of candidate sample points using the Latin hypercube sampling method to obtain a candidate sample set; the candidate sample points include random variables and cycle numbers; Step S3: Based on the candidate sample set, obtain the predicted damage parameter values ​​and the standard deviation of the predicted damage parameter values ​​for each candidate sample point in the candidate sample set through the current data proxy model; Step S4: Calculate the feedback function value of each candidate sample point in the candidate sample set using the feedback function to obtain the candidate feedback function value set; Step S5: If the largest feedback function value in the candidate feedback function value set is greater than the third preset threshold, then perform high-temperature rotating component physical simulation on the candidate sample point corresponding to the largest feedback function value to obtain the corresponding damage parameter benchmark value; form a triplet data by combining the candidate sample point corresponding to the largest feedback function value and the corresponding damage parameter benchmark value, add it to the initial training dataset, and return to step S1; Step S6: If the largest feedback function value in the candidate feedback function value set is less than or equal to the third preset threshold, then the current data proxy model is used as the trained data proxy model. Based on the predicted damage parameters, the predicted reliability of the high-temperature rotating component under the target cycle number is obtained through the damage-threshold interference model; the predicted reliability is the probability that the accumulated damage is less than the damage threshold.

2. The method for predicting the dynamic reliability of high-temperature rotating components based on digital analog driving according to claim 1, characterized in that, Based on the predicted damage parameters, the predicted reliability of the high-temperature rotating component under the target cycle number is obtained through a damage-threshold interference model, specifically including: The prediction reliability of the high-temperature rotating component under the target cycle number is obtained using the following damage-threshold interference model: ； in, This indicates the predicted reliability of a high-temperature rotating component at cycle N. The cumulative distribution function representing the standard normal distribution; Indicates the predicted value of damage parameters The following reliability indicators; Indicates the predicted value of damage parameters The probability density function.

3. The method for predicting the dynamic reliability of high-temperature rotating components based on digital analog driving according to claim 2, characterized in that, The formula for calculating the reliability index is: ； Where A represents the reciprocal of the slope of the linear function of the high-temperature rotating component damage parameters and the mean life obtained by fitting experimental data; A1 represents the reciprocal of the slope of the linear function of the high-temperature rotating component damage parameters and the lower limit of the one-sided standard deviation life obtained by fitting experimental data.

4. The method for predicting the dynamic reliability of high-temperature rotating components based on digital analog driving according to claim 1, characterized in that, The feedback function is: ； ； in, Represents candidate sample points The feedback function value; and represents the predicted damage parameter value and the standard deviation of the predicted damage parameter value for the candidate sample point, respectively; The probability density function representing the standard normal distribution; The cumulative distribution function representing the standard normal distribution; and These represent the upper and lower bounds of standardization, respectively. and These represent the maximum and minimum baseline values ​​of the damage parameter in the initial training dataset, respectively.

5. The method for predicting the dynamic reliability of high-temperature rotating components based on digital analog driving according to claim 1, characterized in that, Data proxy models include artificial neural network models, radial basis function models, extreme learning machine models, support vector machine models, or Gaussian process regression models.

6. A dynamic reliability prediction device for high-temperature rotating components based on digital analog driving, characterized in that, The high-temperature rotating component dynamic reliability prediction device based on digital analog drive applies the high-temperature rotating component dynamic reliability prediction method based on digital analog drive according to any one of claims 1-5, and the high-temperature rotating component dynamic reliability prediction device based on digital analog drive includes: The data acquisition module is used to acquire the random variables and target cycle number of the high-temperature rotating component to be predicted; the random variables include material property parameters, geometric dimension parameters, and load parameters. The model building module is used to construct a damage-threshold interference model based on the interference theory that both damage accumulation and damage threshold follow a normal distribution. The damage prediction module is used to input the random variables of the high-temperature rotating component to be predicted and the target cycle number into a trained data proxy model to obtain the predicted damage parameters and the standard deviation of the predicted damage parameters of the high-temperature rotating component at the target cycle number. The data proxy model is obtained by iteratively training a preset machine learning proxy model based on an initial training dataset until a preset termination condition is met. The initial training dataset includes multiple sets of random variable-cycle number-damage parameter baseline triplet data. The preset termination condition includes the value of the feedback function being less than a preset threshold. The feedback function is constructed based on the predicted damage parameters and the standard deviation of the predicted damage parameters. The reliability acquisition module is used to obtain the predicted reliability of the high-temperature rotating component under the target cycle number based on the predicted damage parameter value and through the damage-threshold interference model; the predicted reliability is the probability that the accumulated damage is less than the damage threshold.

Images