Adaptive generation type number-real fusion test method for aero-engine

Abstract

The invention discloses a self-adaptive generation type number-real fusion test method for an aero-engine. The method comprises the following steps: firstly, establishing a physical entity and a digital twinborn model of an aero-engine key structure, and collecting actual measurement response data in real time for dynamically correcting the digital twinborn model; on the basis, an adaptive sampling strategy is adopted, indexes such as level set uncertainty, model prediction dispersity and expected confidence interval shrinkage are calculated according to conditional distribution information output by the CVAE, and test working conditions with the highest information value are selected batch by batch; then training a CVAE model by using a small amount of measured data and a simulation result, generating a virtual sample covering extreme and complex working conditions, and obtaining a confidence interval of a structure failure probability and a reliability index by combining a weighted statistical inference or Bayesian calibration method; according to the method, a high-confidence reliability result can be obtained under extremely few test conditions, the cost of the reliability test of the aero-engine structure is remarkably reduced, and the evaluation precision and the real-time performance are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of aero-engine structural reliability evaluation, and particularly relates to a self-adaptive generating type numerical-real fusion test method for an aero-engine. BACKGROUND

[0002] Rotary components such as a wheel disc, a blade, a shaft and the like of an aero-engine usually face a severe high-pressure, high-rotation speed and complex gas-machine-vibration coupling load environment in a long-term service process. Long-term alternating load causes structural fatigue, cracks and damage accumulation, and directly affects the safety and reliability of the engine. Traditional reliability evaluation methods mainly rely on physical tests or numerical simulations. Although physical tests can obtain real structural response data, they generally have problems such as high test cost, long test cycle and low effective information value, and it is particularly difficult to adaptively and efficiently explore the performance boundary of the real structure under safety constraints; while numerical simulation analysis (such as finite element analysis) reduces the test cost to a certain extent, but there are parameter uncertainties and systematic errors in the model, resulting in a significant deviation between the predicted results and the actual responses. In addition, the traditional data-driven surrogate models commonly used at present, such as the response surface method or the Kriging model, usually focus on overall trend fitting and lack accurate depiction of response uncertainty near the structural failure boundary, and it is difficult to adaptively and accurately explore the structural performance limit. Therefore, it is urgent to develop a numerical-real fusion test method oriented to adaptive sampling, to quickly and accurately determine the reliability and performance boundary of the rotary component structure of the aero-engine by adaptively and efficiently selecting test conditions and samples and cooperating with accurate modeling of a new generating type model. SUMMARY

[0003] In view of the problems of high reliability test cost, scarcity of critical working conditions and difficulty of accurately positioning the structural failure boundary for the existing aero-engine rotary components, the application proposes an aero-engine self-adaptive generating type numerical-real fusion test method based on adaptive sampling and a conditional variational autoencoder (CVAE). The method of the application firstly constructs a component physical entity and a digital twin model and defines a structural limit state function; then utilizes an adaptive sampling strategy to dynamically optimize the selection of each batch of test conditions according to the real-time model prediction error and response uncertainty; then generates extended structural response virtual samples by fusing a small amount of measured data and digital twin simulation data based on a CVAE conditional probability distribution model; then calibrates the structural failure probability, structural reliability index and confidence interval by using a weighted statistical inference or a Bayesian method; and finally accurately and quickly approximates the real structural failure boundary by sequential adaptive testing, real-time model updating and dynamic confidence interval evaluation, to realize efficient and accurate evaluation of the structural reliability of the aero-engine rotary components.

[0004] To achieve the above object, the application adopts the following technical solutions:

[0005] In a first aspect, the application provides an aero-engine adaptive generation type numerical-real fusion test method, comprising:

[0006] Step S1, based on the actual structure size, material properties and typical task load spectrum of the rotating parts of the aero-engine, a physical finite element model and a corresponding digital twin model of the rotating parts are constructed, and a structure limit state function for reliability evaluation is determined;

[0007] Step S2, initial simulation analysis is performed using the digital twin model, and parameter sensitivity analysis and prediction response error estimation are combined to preliminarily identify the limit state neighborhood close to the structure failure boundary, and the first batch of test working conditions is determined;

[0008] Step S3, according to the first batch of test working conditions, the first batch of tests are carried out, the measured response data of the key structure parts are collected in real time and synchronously, the measured response data are compared with the prediction results of the digital twin model, and the model prediction error and local response uncertainty are quantitatively analyzed; the measured response data include strain, stress and vibration of the key structure parts;

[0009] Step S4, an adaptive sampling strategy is adopted, the boundary proximity, local response uncertainty, model prediction error and estimated structure reliability index accuracy improvement potential of the current digital twin model prediction are comprehensively considered, and information value scores of candidate working conditions are obtained;

[0010] Step S5, according to the information value score results of step S4, the sampling range and parameter distribution of the next batch of test working conditions are adaptively determined under the premise of meeting the safety margin, and the next batch of tests are carried out according to the next batch of test working conditions; the next batch of test working conditions are used to accurately and efficiently approach the real failure boundary of the structure;

[0011] Step S6, a conditional variational autoencoder is used as a generative model, the measured response data obtained in the test and the prediction results of the digital twin model are used as the training basis, a conditional probability distribution model of the structure response is trained, and a structure response virtual sample is generated; the structure response virtual sample is used to cover a wider range of working conditions, including complex working conditions and extreme working conditions that are not actually measured;

[0012] Step S7, the structure response virtual sample and the measured response data are fused, a weighted statistical inference analysis is carried out, the structure failure probability and the structure reliability index are quantitatively calculated, the confidence interval of the structure reliability index is given by combining the confidence interval estimation or the Bayesian calibration method, and the uncertainty of the statistical estimation is evaluated;

[0013] Step S8, according to the structure reliability index and the confidence interval thereof obtained in real time, it is judged whether the structure reliability index meets the preset convergence condition; if the structure reliability index does not meet the preset convergence condition, the adaptive test loop is repeatedly executed by returning to executing steps S4 to S7, the test working condition selection strategy, the generating model parameter and the reliability estimation strategy are dynamically adjusted, and the real structure failure boundary is gradually and accurately and efficiently approximated;

[0014] Step S9, when the structure reliability index meets the preset convergence condition, the test loop is terminated, and the final structure failure probability, the structure reliability index and the confidence interval thereof are output, and the accurate evaluation of the structure reliability of the aero-engine rotating component is completed.

[0015] In a second aspect, the present application provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the aforementioned adaptive generating-numerical-real fusion test method of aero-engine.

[0016] The beneficial effects of the present application are:

[0017] (1) Sequential test decision driven by adaptive sampling: a test value scoring mechanism is proposed, which takes boundary proximity, model prediction error, confidence interval contraction and effective sample size gain as the core, a small number of working conditions are selected in batches, and the criteria and model are updated immediately after each batch, the efficient approximation and information efficiency maximization of the failure boundary are realized, and the number and period of real tests are significantly reduced.

[0018] (2) Conditional generating probability modeling enhances reliability inference: a conditional variational autoencoder is used to construct the conditional distribution of "working condition-response", combined with Bayesian calibration and safety quantile constraint, to generate compliant samples for candidate evaluation, missing data completion and proposal sampling; under the inference framework with important weight, the reviewable failure probability and reliability interval are output, and the robustness and extrapolation ability under complex working conditions and small sample conditions are improved.

[0019] (3) Numerical-real closed loop assimilation and variance reduction: taking digital twin as the physical prior and control variable channel, combining with the generating model of small step incremental update, forming a closed loop process of "test-assimilation-generation-inference"; under the premise of meeting the equipment red line and safety constraint, the reliability index is quickly converged and the statistical variance and calculation overhead are reduced, and the engineering feasibility and traceability are considered. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The flowchart of the adaptive generating-numerical-real fusion test method of aero-engine of the present application. DETAILED DESCRIPTION

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0023] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0024] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0025] Furthermore, to better illustrate the present invention, numerous specific details are provided in the following detailed embodiments. Those skilled in the art should understand that the present invention can be practiced without certain specific details. In some instances, methods, means, elements, and circuits well known to those skilled in the art have not been described in detail in order to highlight the spirit of the invention.

[0026] Figure 1 A flowchart illustrating the adaptive generative numerical-real fusion test method for aero-engines of the present invention is shown, as follows: Figure 1 As shown, the method includes:

[0027] Step S1: Based on the actual structural dimensions, material properties, and typical mission load spectrum of the rotating components of the aero-engine, construct the physical finite element model and the corresponding digital twin model of the rotating components, and determine the structural limit state function used for reliability assessment.

[0028] For example, step S1 may include:

[0029] Step S11: Establish the structural limit state function of the rotating component to describe the safety margin of the rotating component under multi-field coupling conditions. The structural limit state function is defined as follows:

[0030] (1)

[0031] in, The structure of the rotating component under operating conditions The performance margin is as follows; The structural response variable of the rotating component. The equivalent criteria (e.g., equivalent stress, equivalent strain, or deformation) are obtained by mapping the structural response variables. These are the corresponding allowable limits used to determine whether the rotating component has failed. It can be configured by those skilled in the art according to actual needs. When If the rotating component fails, it is determined that the component has failed.

[0032] Step S12: Establish the digital twin model to predict the mapping relationship between the working condition input and the structural response output. The model expression is:

[0033] (2)

[0034] in, For digital twin models in working conditions The predicted response of the output; The simulation response function is based on finite element method and physical information. This is the set of parameters to be calibrated for the digital twin model.

[0035] Step S2: Use the digital twin model to perform initial simulation analysis, and combine parameter sensitivity analysis and predicted response error estimation to initially identify the limit state neighborhood close to the structural failure boundary and determine a small number of initial test conditions.

[0036] For example, after initial simulation analysis, sensitivity analysis can be used to determine the degree of influence of key operating parameters on the structural limit state and to screen high-risk areas; the initial limit state neighborhood is defined as follows:

[0037] (3)

[0038] in, This is the initial set of operating conditions approaching the structural failure boundary; For digital twin models in working conditions The predicted structural limit state function value; The threshold for determining the boundary neighborhood can be set by those skilled in the art according to actual needs. That is, the high-risk areas identified through screening, within the region The first batch of test conditions are randomly selected or based on the weight of the test condition spectrum, and the first batch of tests are carried out. Key response data are collected as input for subsequent model updates and validation.

[0039] Step S3: Conduct the first batch of tests according to the first batch of test conditions, collect the measured response data such as strain, stress, and vibration of key structural parts in real time, compare the measured response data with the prediction results of the digital twin model, and quantitatively analyze the model prediction error and local response uncertainty.

[0040] For example, step S3 may include:

[0041] Step S31: Compare the predicted response output by the digital twin model with the measured response data, and define the model difference degree to quantitatively characterize the model prediction error. The model difference degree is defined as follows:

[0042] (4)

[0043] in, For working conditions Model difference; For working conditions The measured response under the given conditions; To measure the noise covariance matrix; This indicates transpose.

[0044] Step S32: Calculate the local response uncertainty based on the measured response data to reflect the influence of measurement fluctuations and environmental disturbances. The local response uncertainty is:

[0045] (5)

[0046] in, For working conditions Local variance; For sliding time windows; The mean of the limit state function of the structure within the window; for Real-time data collection of operating conditions The measured response under the given conditions This is the instantaneous performance margin value calculated from the measured response using the structural limit state function.

[0047] Step S4: Adopt an adaptive sampling strategy, comprehensively consider the boundary proximity of the current digital twin model prediction, local response uncertainty, model prediction error, and the potential for improving the accuracy of the estimated structural reliability index, and score the information value of the candidate working conditions.

[0048] For example, step S4 may include:

[0049] Step S41: Estimate the level set uncertainty of structural failure using the conditional distribution to evaluate the information value of the test conditions. The level set uncertainty is:

[0050] (6)

[0051] in, Indicates working conditions The level set uncertainty is used to measure the uncertainty of the location of the structural failure boundary; For working conditions The expected value of the lower structure limit state; For working conditions The standard deviation of the limit state of the lower structure; The balancing coefficient can be set by those skilled in the art according to actual needs.

[0052] Step S42: Taking into account both model prediction error and information value, the adaptive sampling scoring function is defined as follows:

[0053] (7)

[0054] in, For working conditions The overall score; For working conditions The mutual information between the digital twin model before and after the update is used to measure the information gain of the model between the current prediction and subsequent updates. For working conditions The expected confidence interval contraction value; For working conditions The model difference term before and after the digital twin model update is used to reflect the degree of deviation between the model prediction and the measured response; , , , These are adjustable weighting coefficients.

[0055] Step S5: Based on the information value scoring results of step S4, and under the premise of meeting the safety margin, adaptively determine the sampling range and parameter distribution of the next batch of test conditions, and carry out the next batch of tests based on the next batch of test conditions; the next batch of test conditions is used to accurately and efficiently approximate the actual failure boundary of the structure.

[0056] For example, it can be based on an adaptive sampling scoring function. The candidate operating conditions are ranked, and the one with the most information is selected under the safety quantile constraint. This operating condition will be used as the next batch of test operating conditions:

[0057] (8)

[0058] in, This is the set of selected test conditions for the next batch of trials. For the set of candidate operating conditions; A candidate operating condition library; For working conditions The confidence quantile can be set, for example, to a 95% safety limit. Based on the determined next batch of test conditions, conduct the next batch of tests and collect measured response data to complete a new round of structural response observations, providing input for updating the generative model and digital twin model.

[0059] Step S6: Using CVAE as a generative model, and based on the measured response data obtained in the experiment and the prediction results of the digital twin model, a conditional probability distribution model of the structural response is trained, and a large number of virtual samples of structural response that can cover a wider range of working conditions (especially complex and extreme working conditions that have not been actually measured) are generated.

[0060] For example, step S6 may include:

[0061] Step S61: Using CVAE as a generative model, model the conditional probability distribution of the structural response. The model structure is defined as follows:

[0062] (9)

[0063] in, For working conditions The structural response output is as follows; These are latent variables in generative models, used to characterize unobserved latent features or noise perturbations; For generative models in working conditions The output response correction amount; For conditional latent variable distribution; For digital twin models in working conditions The predicted response.

[0064] Step S62: Train the generative model using a boundary-weighted variational objective function to enhance its generative capability under boundary conditions.

[0065] (10)

[0066] in, Let be the weighted variational objective function; For the first The sample weights of each sample; For the first Input of working conditions for each sample; For the first The structural response output corresponding to each sample; For the distribution of conditional latent variables Expectations; The conditional probability distribution predicted by the generative model; For digital twin models in working conditions Predicted response; In working conditions With latent variables Conditional inputs generated under the combined effect; The regularization coefficient; This represents the Kullback-Leibler (KL) divergence. for and KL divergence between them; The posterior distribution of latent variables estimated for generative models is an approximation of the true distribution of conditional latent variables; Latent variables The prior distribution of .

[0067] Step S7: Fuse the virtual sample of the structural response with the measured response data, perform weighted statistical inference analysis, quantitatively calculate the structural failure probability and structural reliability index, and combine confidence interval estimation or Bayesian calibration method to give the precise confidence interval of the structural reliability index and evaluate the uncertainty of statistical estimation.

[0068] For example, step S7 may include:

[0069] Step S71: Combining the virtual structural response samples and the measured response samples, calculate the structural failure probability using a weighted importance estimation method. :

[0070] (11)

[0071] in, According to the first Structural response output of each sample The calculated limit state function value; For the first The working conditions corresponding to each sample; For the working condition spectrum probability; The sample point distribution is used to describe the probability density of each working condition being selected in adaptive sampling. The total number of samples used for failure probability calculation includes generated virtual structural response samples and measured response samples. The two types of samples are processed uniformly according to their respective sample weights during the weighted statistical process, without any fixed limit on the proportion; the measured response samples are obtained from the measured response data.

[0072] Step S72: Calculate the structural reliability index based on the structural failure probability. :

[0073] (12)

[0074] in, It is the inverse function of the standard normal distribution, used to express the probability of structural failure. Converted into structural reliability index Structural reliability indicators are used to reflect the confidence level of structural safety margin.

[0075] Step S8: Based on the real-time obtained structural reliability index and its confidence interval, determine whether the structural reliability index meets the preset convergence condition; if the structural reliability index does not meet the preset convergence condition, return to steps S4 to S7, repeat the adaptive test loop, dynamically adjust the test condition selection strategy, generative model parameters and reliability estimation strategy, and gradually and accurately and efficiently approach the real structural failure boundary.

[0076] For example, the confidence interval width of the structural reliability index can be monitored in real time, and the structural reliability index can be judged to meet the preset convergence condition based on the confidence interval width or the structural failure probability. The preset convergence condition (also known as the adaptive convergence criterion) is as follows:

[0077] (13)

[0078] in, The confidence interval width of the structural reliability index is used to measure the convergence of the structural reliability index estimation results. Estimated structural failure probability The coefficient of variation is used to reflect the uncertainty of the failure probability estimate; This is the confidence interval convergence threshold; The convergence threshold for the coefficient of variation; and This can be set by those skilled in the art according to actual needs. When the structural reliability index fails to converge, steps S4 to S7 are repeated, and the newly added data is used for incremental updates of the parameters of the digital twin model and the generative model until the preset convergence condition is met.

[0079] Step S9: When the structural reliability index meets the preset convergence condition, terminate the test cycle, output the final structural failure probability, structural reliability index and its confidence interval, and complete the accurate assessment of the structural reliability of the aero-engine rotating component.

[0080] This invention addresses the challenges of complex operating conditions and high testing costs associated with rotating components by integrating adaptive sampling with CVAE (Continuous Cell Image Processing) to propose an adaptive generative modeling strategy. By establishing a virtual mapping relationship between the physical entity and digital twin model of a key aero-engine structure, strain, stress, and vibration data are collected in real time to dynamically correct the digital twin model. Based on this, an adaptive sampling strategy is employed to calculate indicators such as level set uncertainty, model prediction dispersion, and expected confidence interval shrinkage according to the conditional distribution information output by CVAE. The most informative test conditions are selected batch by batch to efficiently explore the critical boundaries of aero-engines. The CVAE model is trained using a small amount of measured data and simulation results to generate virtual samples of structural responses covering extreme and complex conditions. These samples are then combined with weighted statistical inference or Bayesian calibration methods to obtain confidence intervals for structural failure probability and reliability indicators. This invention achieves a closed-loop adaptive update of the "test-modeling-evaluation" process, obtaining high-confidence reliability results with minimal testing conditions. This significantly reduces the cost of aero-engine structural reliability testing and improves the accuracy and real-time performance of the evaluation.

[0081] The following describes in detail the specific implementation process of the adaptive generative numerical-real fusion test method for aero-engines of the present invention, using a certain type of aero-engine fan-compressor system as a specific example. The steps of the adaptive generative numerical-real fusion test method for aero-engines in this embodiment include:

[0082] Step S1: Based on the actual geometric structure and material parameters of the fan-compressor system, establish its high-precision finite element physical model and corresponding digital twin model, clarify the mapping relationship of key positions such as the fan blade tip and the compressor disk rim, and define the limit state function of the system structure according to the typical flight condition spectrum of the engine to describe the safety margin of the structure under different operating conditions.

[0083] Step S2: Using the established digital twin model, conduct preliminary simulation analysis of the fan-compressor system under typical flight conditions such as cruise, takeoff and climb. Through parameter sensitivity analysis and prediction of response error, identify the high-risk areas near the blade tip crack initiation and the disk rim fatigue crack propagation, and determine the key operating parameters for the first batch of tests.

[0084] Step S3: Conduct tests based on the key operating parameters determined in Step S2. Install high-precision stress, strain, and vibration sensors at key locations such as the fan blade tip and compressor disk rim. Collect the structural response data of the system under typical flight conditions in real time and synchronously. Compare the measured data with the predicted response of the digital twin model one by one to quantitatively analyze the model prediction error and the uncertainty of local response.

[0085] Step S4: Based on the adaptive sampling strategy, comprehensively consider the boundary proximity of the current digital twin model prediction, local response uncertainty, model prediction error, and the potential for improving the accuracy of reliability indicators, calculate the information value score of the candidate working conditions, and adaptively determine the sampling range and parameter distribution of the next batch of test working conditions.

[0086] Step S5: Based on the scoring results of step S4, and under the premise of satisfying the safety margin of the fan-compressor system, adaptively select the next batch of test conditions to accurately and efficiently approximate the true failure boundary of fan blade tip crack initiation and disc rim crack propagation.

[0087] Step S6: Using a conditional variational autoencoder as a generative model, train a conditional probability distribution model with the limited experimental data and simulation results of the digital twin model to efficiently generate virtual structural response samples covering a wider range of working conditions (especially complex and extreme working conditions that have not yet been actually measured) to supplement the lack of real experimental data.

[0088] Step S7: The generated virtual structural response samples are fused with the actual experimental data. A weighted statistical inference method is used to quantitatively calculate the failure probability and reliability index of key components of the fan-compressor system. The confidence interval of the reliability index is accurately given by confidence interval estimation or Bayesian calibration method, and the uncertainty of the statistical estimation is evaluated.

[0089] Step S8: Based on the real-time obtained reliability indicators and their confidence intervals, determine whether the preset convergence conditions are met; if not, repeat steps S4 to S7 to dynamically adjust the test condition selection, generative model parameters and reliability assessment strategies to gradually and accurately explore the real failure boundary.

[0090] Step S9: When the reliability index meets the preset convergence condition, terminate the test cycle and output the final structural failure probability, reliability index and corresponding confidence limit of the fan-compressor system, so as to realize the accurate assessment of the structural reliability of the aero-engine fan-compressor system.

[0091] The present invention also proposes an electronic device, comprising: one or more processors; and a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the above-described adaptive generative data-real fusion test method for aero-engines.

[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0093] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0097] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. The above descriptions are exemplary and not exhaustive. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive generative numerical-real fusion test method for aero-engines, characterized in that, include: Step S1: Based on the actual structural dimensions, material properties, and typical mission load spectrum of the rotating components of the aero-engine, construct the physical finite element model and the corresponding digital twin model of the rotating components, and determine the structural limit state function used for reliability assessment. Step S2: Use the digital twin model to perform initial simulation analysis, and combine parameter sensitivity analysis and predicted response error estimation to initially identify the limit state neighborhood close to the structural failure boundary and determine the first batch of test conditions; Step S3: Conduct the first batch of tests according to the first batch of test conditions, collect the measured response data of key structural parts in real time, compare the measured response data with the prediction results of the digital twin model, and quantitatively analyze the model prediction error and local response uncertainty; the measured response data includes strain, stress and vibration of key structural parts; Step S4: Adopt an adaptive sampling strategy, comprehensively consider the boundary proximity of the current digital twin model prediction, local response uncertainty, model prediction error, and the potential for improving the accuracy of the estimated structural reliability index, and score the information value of the candidate working conditions. Step S5: Based on the information value scoring results of step S4, and under the premise of meeting the safety margin, adaptively determine the sampling range and parameter distribution of the next batch of test conditions, and carry out the next batch of tests according to the next batch of test conditions; the next batch of test conditions is used to accurately and efficiently approximate the actual failure boundary of the structure. Step S6: Using a conditional variational autoencoder as a generative model, and based on the measured response data obtained in the experiment and the prediction results of the digital twin model, a conditional probability distribution model of the structural response is trained, and virtual samples of the structural response are generated. The virtual structural response samples are used to cover a wider range of operating conditions, including complex and extreme conditions that have not been actually measured. Step S7: Fuse the virtual sample of the structural response with the measured response data, perform weighted statistical inference analysis, quantitatively calculate the structural failure probability and structural reliability index, and combine confidence interval estimation or Bayesian calibration method to give the confidence interval of the structural reliability index and evaluate the uncertainty of statistical estimation. Step S8: Based on the real-time obtained structural reliability index and its confidence interval, determine whether the structural reliability index meets the preset convergence condition. If the structural reliability index does not meet the preset convergence condition, return to steps S4 to S7 and repeat the adaptive test cycle, dynamically adjust the test condition selection strategy, generative model parameters and reliability estimation strategy, and gradually and accurately and efficiently approach the real structural failure boundary. Step S9: When the structural reliability index meets the preset convergence condition, terminate the test cycle, output the final structural failure probability, structural reliability index and its confidence interval, and complete the accurate assessment of the structural reliability of the aero-engine rotating component.

2. The method according to claim 1, characterized in that, Step S1 includes: Step S11: Establish the structural limit state function of the rotating component to describe the safety margin of the rotating component under multi-field coupling conditions. The structural limit state function is defined as follows: (1) in, The structure of the rotating component under operating conditions Performance margin below; The structural response variable of the rotating component. This is the equivalent criterion obtained by mapping the structural response variables; The corresponding allowable limit is used to determine whether the rotating component has failed; when When this occurs, the rotating component is determined to be faulty; Step S12: Establish the digital twin model to predict the mapping relationship between the working condition input and the structural response output. The model expression is: (2) in, For the digital twin model in working conditions The predicted response of the output; The simulation response function is based on finite element method and physical information. This is the set of parameters to be calibrated for the digital twin model.

3. The method according to claim 2, characterized in that, In step S2, after initial simulation analysis, sensitivity analysis is used to determine the degree of influence of key operating parameters on the structural limit state, and high-risk areas are screened; the initial limit state neighborhood is defined as follows: (3) in, This is the initial set of operating conditions approaching the structural failure boundary; For the digital twin model in working conditions The predicted structural limit state function value; The threshold for determining the boundary neighborhood; In the set The first batch of test conditions are selected randomly or based on the weighting of the test condition spectrum.

4. The method according to claim 3, characterized in that, Step S3 includes: Step S31: Compare the predicted response output by the digital twin model with the measured response data, and define the model difference degree to quantitatively characterize the model prediction error. The model difference degree is defined as follows: (4) in, For working conditions Model difference; For working conditions The measured response under the given conditions; To measure the noise covariance matrix; Indicates transpose; Step S32: Calculate the local response uncertainty based on the measured response data to reflect the influence of measurement fluctuations and environmental disturbances. The local response uncertainty is: (5) in, For working conditions Local variance; For sliding time windows; The mean of the limit state function of the structure within the window; for Real-time data collection of operating conditions The measured response under the given conditions This is the instantaneous performance margin value calculated from the measured response using the structural limit state function.

5. The method according to claim 4, characterized in that, Step S4 includes: Step S41: Estimate the level set uncertainty of structural failure using the conditional distribution to evaluate the information value of the test conditions. The level set uncertainty is: (6) in, For working conditions The level set uncertainty is used to measure the uncertainty of the location of the structural failure boundary; For working conditions The expected value of the lower structure limit state; For working conditions The standard deviation of the limit state of the lower structure; This is the balance coefficient; Step S42: Taking into account both model prediction error and information value, the adaptive sampling scoring function is defined as follows: (7) in, For working conditions The overall score; For working conditions The mutual information between models is used to measure the information gain of the model between the current prediction and subsequent updates. For working conditions The expected confidence interval contraction value; For working conditions The model difference term below is used to reflect the degree of deviation between the model prediction and the measured response; , , , These are adjustable weighting coefficients.

6. The method according to claim 5, characterized in that, In step S5, according to the adaptive sampling scoring function The candidate operating conditions are ranked, and the one with the most information is selected under the safety quantile constraint. This operating condition will be used as the next batch of test operating conditions: (8) in, This is the set of selected test conditions for the next batch of trials. For the set of candidate operating conditions; A candidate operating condition library; For working conditions The confidence quantiles below; The next batch of tests will be conducted based on the next batch of test conditions, and measured response data will be collected to complete a new round of structural response observation, providing input for the updating of the generative model and the digital twin model.

7. The method according to claim 6, characterized in that, Step S6 includes: Step S61: Using a conditional variational autoencoder as a generative model, model the conditional probability distribution of the structural response. The model structure is defined as follows: (9) in, For working conditions The structural response output is as follows; These are the latent variables in the generative model, used to characterize unobserved latent features or noise perturbations; For the generative model in the working condition The output response correction amount; For conditional latent variable distribution; For the digital twin model in working conditions Predicted response; Step S62: Train the generative model using a boundary-weighted variational objective function to enhance its generative capability under boundary conditions. (10) in, Let be the weighted variational objective function; For the first The sample weights of each sample; For the first Input of working conditions for each sample; For the first The structural response output corresponding to each sample; For the distribution of conditional latent variables Expectations; The conditional probability distribution predicted by the generative model; For the digital twin model in working conditions Predicted response; In working conditions With latent variables Conditional inputs generated under the combined effect; The regularization coefficient; Kullback-Leibler divergence; The posterior distribution of the latent variables estimated for the generative model is an approximation of the true distribution of the conditional latent variables; Latent variables The prior distribution of .

8. The method according to claim 7, characterized in that, Step S7 includes: Step S71: Combining the virtual structural response samples and the measured response samples, calculate the structural failure probability using a weighted importance estimation method. : (11) in, According to the first Structural response output of each sample The calculated limit state function value; For the working condition spectrum probability; The sample point distribution is used to describe the probability density of each working condition being selected in adaptive sampling. The sample size; the measured response samples are obtained from the measured response data; Step S72: Calculate the structural reliability index based on the structural failure probability. : (12) in, It is the inverse function of the standard normal distribution, used to express the probability of structural failure. Converted into structural reliability index Structural reliability indicators are used to reflect the confidence level of structural safety margin.

9. The method according to claim 8, characterized in that, In step S8, the confidence interval width of the structural reliability index is monitored in real time. Based on the confidence interval width or the structural failure probability, it is determined whether the structural reliability index meets the preset convergence condition. The preset convergence condition is: (13) in, The confidence interval width of the structural reliability index is used to measure the convergence of the structural reliability index estimation results. Structural failure probability The coefficient of variation is used to reflect the uncertainty of the failure probability estimate; This is the confidence interval convergence threshold; The convergence threshold for the coefficient of variation; If the structural reliability index fails to converge, repeat steps S4 to S7 and use the new data for incremental updates of the model parameters until the preset convergence condition is met.

10. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more programs are executed by the one or more processors, the one or more processors implement the adaptive generative data-real fusion test method for aero-engines as described in any one of claims 1-9.

Images