A Distributed Collaborative Reliability Method for Multi-Failure Structures Based on Discipline Decomposition

Abstract

The present invention discloses a distributed collaborative reliability method for a multi-failure structure based on subject decomposition, which is applied to the field of aircraft engine reliability technology. The method comprises the following steps: first, the various failure modes of the multi-failure structure are disassembled according to the subjects to which they belong, the response analysis of a single subject is carried out, and a distributed proxy model of the response of each subject is established; then, the response of each subject is extracted in linkage by using the collaborative sampling technology, and the distributed collaborative reliability analysis of the multi-failure structure is completed by using the established distributed proxy model; finally, the reliability analysis results of the multi-failure structure are imported into the Copula correlation model, and the correlation between multiple failure modes is evaluated by simulation results. Compared with the prior art, the present invention combines the distributed collaborative proxy model based on subject decomposition with the Copula function that can quantitatively quantify the correlation, which not only accurately quantifies the failure correlation between each failure mode, but also ensures high calculation accuracy and efficiency.

Description

Technical Field

[0001] The present invention is a reliability method for a multi-failure structure containing multiple failure modes, and in particular relates to a distributed collaborative reliability method for a multi-failure structure based on subject decomposition. The method can consider the correlation between multiple failure modes and belongs to the field of aircraft engine reliability technology. Background Art

[0002] Multi-failure structures such as turbine blades will cause multiple failure modes such as low-cycle fatigue, high-cycle fatigue, and creep under the coupling of multi-physical field loads such as flow, solid, and heat. The design analysis problem is a complex design analysis problem involving multiple functional functions and multi-disciplinary responses. For example, in the structure of aero-engine turbine blades, the establishment of a low-cycle fatigue functional function based on strain fatigue characteristics requires a pseudo-static response analysis for the working state; while the establishment of a high-cycle fatigue functional function based on stress fatigue characteristics requires a dynamic response analysis for structural vibration. However, the collaborative analysis of the deterministic multi-disciplinary response of multi-failure structures is already cumbersome. On this basis, the computational efficiency and accuracy of the multi-disciplinary collaborative response probabilistic analysis will become a bottleneck restricting its engineering application. The surrogate model method is a common method to improve the computational efficiency of the probabilistic analysis of multi-failure structures. At present, this method has been widely used in the probabilistic analysis of general engineering structures and has shown certain superiority. However, the functional functions corresponding to the multiple failure modes of multi-failure structures often belong to different disciplines. Therefore, it is necessary to conduct a multi-disciplinary probabilistic analysis in the reliability analysis of multi-failure structures. If an overall proxy model is established directly, it is difficult for the established proxy model to accurately reflect the complex mapping relationship between multidisciplinary responses and random variables, and its calculation accuracy cannot be guaranteed; if interactive calculations of various disciplines are performed when establishing the proxy model, its calculation efficiency is very low and has no practical engineering value, so that the traditional proxy model method also faces severe challenges in calculation accuracy and efficiency. Summary of the invention

[0003] The technical problem to be solved by the present invention is: to overcome the deficiencies of the prior art, to provide a distributed collaborative reliability method for multi-failure structures taking into account the correlation of multi-mode failures, and to improve the calculation accuracy and efficiency of the reliability of multi-failure structures of aircraft engines.

[0004] The technical solution adopted by the present invention is: a multi-failure structure distributed collaborative reliability method based on subject decomposition, comprising:

[0005] Step S1, for a multi-failure structure including multiple failure modes, the types and quantities of failure modes of the multi-failure structure are counted according to actual engineering practice; the multi-failure structure including multiple failure modes refers to a complex engineering structure that produces multiple failure modes under the action of complex loads such as fluid load, thermal load, and solid load, and the multiple failure modes include: low-cycle fatigue failure, high-cycle fatigue failure, and creep-fatigue failure;

[0006] Step S2, disassemble the multiple failure modes according to the different disciplines to which they belong, conduct response analysis of a single discipline respectively, and obtain the output response of a single discipline, that is, the response of each discipline; the disciplines include fluid mechanics, thermodynamics, solid mechanics, etc.;

[0007] Step S3, according to the characteristics of each failure mode, respectively obtain the input random variable and output response of each failure mode, and use the input random variable and output response of each failure mode as the input parameter and response parameter of the multi-failure structure; the characteristics of each failure mode include that low-cycle fatigue failure requires a quasi-static response analysis for state cycles, and cycle fatigue requires a dynamic response analysis for the structure;

[0008] Step S4, performing deterministic analysis on the multi-failure structure, obtaining dangerous parts of the multi-failure structure, completing deterministic finite element analysis, and obtaining deterministic finite element simulation results;

[0009] Step S5, using collaborative sampling technology to extract a small batch of input samples of about 30-200, combined with deterministic finite element simulation results, to obtain a one-to-one corresponding output response of dangerous parts of the multi-failure structure, and establish a database of input parameters and response parameters containing correlation relationships;

[0010] Step S6, establishing different types of distributed agent models for each subject response, fitting the agent model according to the input parameters and response parameter database established in step 5, and selecting the agent model with the highest accuracy as the optimal agent model for each subject response;

[0011] Step S7, establishing a reliability model related to multi-mode failure based on the resistance probability distribution model and the generalized "stress-strength" interference method;

[0012] Step S8, based on the reliability model in step S7, using collaborative sampling technology to extract responses from various disciplines, using the optimal proxy model in step S6 to complete the reliability analysis of the multi-failure structure, and obtaining the reliability of the multi-failure structure;

[0013] Step S9, import the reliability analysis results of the multi-failure structure obtained in step S8 into the Copula correlation model, quantitatively quantify the correlation between the failure modes with the simulation results, and obtain the correlation between the failure modes of the multi-failure structure. Only on the basis of quantitatively evaluating the failure correlation can an accurate structural system reliability model be established, that is, the correlation between the failure modes of the multi-failure structure is obtained, and the reliability of the multi-failure structure is accurately predicted.

[0014] The above invention claims are innovative and have advantages in that: (1) Step S2 decomposes the multiple failure modes of the multi-failure structure according to the subject response. The advantage is that, on the one hand, the proxy model is established according to the analysis of a single subject, which reduces the model size and helps to ensure the accuracy of the model. On the other hand, the decomposition operation of the subject is also conducive to improving the computational efficiency of the reliability analysis of the multi-failure structure; (2) Step S5 The collaborative sampling technology considers the correlation between the input parameters and the response parameters, cleverly solves the availability and synchronization of the extracted samples, thereby ensuring the rationality of multiple design objectives, parameters and models, making it possible to process multiple related output responses at the same time.

[0015] The specific steps of the collaborative sampling technology in step S5 are: using the Latin hypercube sampling method to sample all input parameters of the multi-failure structure and the response parameters of each subject simultaneously for multiple times.

[0016] The above innovation lies in: the collaborative sampling technology takes into account the correlation between input parameters and response parameters, cleverly solves the availability and synchronization of sample extraction, thereby ensuring the rationality of multiple design objectives, parameters and models, making it possible to process multiple related output responses simultaneously.

[0017] When selecting the proxy model in step S6, a proxy model with high accuracy is selected by minimizing the loss function according to the characteristics of the reliability analysis of each subject response.

[0018] Agent model types include polynomial models, kriging models, support vector machine models, machine learning models, and deep learning models.

[0019] The above innovation and advantage lies in selecting different types of agent models according to the response characteristics of different disciplines, which can maximize the potential and advantages of multiple types of agent models while avoiding the limitations of a single type of agent model.

[0020] The step S8 comprises the following sub-steps:

[0021] Step S81, using the optimal proxy model established by each subject response to replace the multi-failure structure to perform multi-mode collaborative probability analysis;

[0022] Step S82: Based on the allowable value of each subject response, that is, judging whether the response of each subject after the multi-failure structure is loaded is too large or too small, a standard of measurement is determined in advance, and a multi-mode limit state equation is established. Assume that the reliability analysis of the multi-failure structure involves l response subjects, and the pth allowable output response is [Y (p) ], then the limit state equation g under the pth corresponding discipline (p) (x) can be expressed as:

[0023] (3)

[0024] in, is the optimal solution obtained by the p-th distributed agent model;

[0025] Step S83, using collaborative sampling technology to perform large-scale sampling of about 10,000-100,000 input parameters, and bringing them into the optimal proxy model to calculate the response parameters corresponding to each failure mode;

[0026] Step S84: determine whether the response value of each subject falls within the safety domain according to the multi-mode limit state equation. The number of sample points falling within the safety domain is counted, and the reliability value is approximately replaced by the frequency of the sample points within the safety domain, that is, the reliability of the multi-failure structure is obtained, and the reliability analysis of the multi-failure structure is completed. The reliability calculation formula is as follows:

[0027] (4)

[0028] The above innovations and advantages lie in: dividing the various failure modes of multi-failure structures into l response disciplines according to the subject response characteristics, thereby converting complex multi-disciplinary problems into a series of sub-problems of a single discipline, reducing the model size, which is beneficial to ensuring the accuracy of the model and improving the computational efficiency of reliability analysis of multi-failure structures.

[0029] The step S9 includes the following sub-steps:

[0030] Step S91, based on the database of input parameters and response parameters containing correlation established in step S5, solve the Copula parameters using the maximum likelihood estimation method;

[0031] Step S92, using AIC and BIC criteria to select a suitable Copula function type, and establish a Copula correlation model for correlation quantification;

[0032] Step S93, importing the reliability analysis result of the multi-failure structure calculated in step S8 into the Copula correlation model to evaluate the correlation between the multi-failure modes.

[0033] The above innovation and advantages lie in that: under the assumption of mutual independence, the effect of failure correlation is underestimated, resulting in a relatively safe reliability design result, while the assumption of complete correlation overestimates the effect of failure correlation, resulting in a relatively dangerous reliability design result, making it impossible for the design to meet the requirements of reliability indicators. The claims of the present invention take into account the correlation between input parameters and response parameters, and use the Copula correlation model to quantify the correlation between multiple failure modes.

[0034] Compared with the prior art, the present invention has the following advantages:

[0035] (1) Discipline decomposition: The various failure modes of multi-failure structures are organically disassembled according to discipline responses. Its advantages are that, on the one hand, since the proxy model is established according to the analysis of a single discipline, the model size is reduced, which is conducive to ensuring the accuracy of the model. On the other hand, the discipline decomposition operation is also conducive to improving the computational efficiency of multi-failure structure reliability analysis, providing an effective way for multi-failure structure reliability analysis;

[0036] (2) Collaborative sampling: It can consider the correlation between input parameters and response parameters, cleverly solve the availability and synchronization of sample extraction, and thus ensure the rationality of multiple design objectives, parameters and models, making it possible to process multiple related output responses simultaneously, which is beneficial to improving the calculation accuracy of reliability analysis of multi-failure structures such as aircraft engines. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flow chart of the multi-failure structure distributed collaborative reliability method based on subject decomposition of the present invention. DETAILED DESCRIPTION

[0038] The present invention will be further described below in conjunction with specific embodiments, but is not limited to the specific embodiments.

[0039] The following takes the high-cycle fatigue and low-cycle fatigue reliability analysis of turbine blades with multiple failure structures as an example to further illustrate the invented distributed collaborative reliability method for multiple failure structures based on subject decomposition. The specific steps are as follows:

[0040] like Figure 1 As shown, the multi-failure structure distributed collaborative reliability method based on subject decomposition of the present invention includes the following steps:

[0041] Step 1: Under the coupling of multi-physical field loads such as fluid load, thermal load, solid load, etc., turbine blades will induce various failure modes such as low-cycle fatigue and high-cycle fatigue.

[0042] Step 2: Disassemble the failure modes according to the different disciplines they belong to, and conduct response analysis of single disciplines such as fluid mechanics, thermodynamics, and solid mechanics.

[0043] Step 3, obtaining the input random variables and output responses of high cycle fatigue and low cycle fatigue of the turbine blade, and using them as input parameters and response parameters of the turbine blade;

[0044] Step 4, by performing deterministic analysis on the turbine blades, it is found that the turbine blade root has low-cycle fatigue and high-cycle fatigue, and the turbine blade root is regarded as a dangerous part to complete deterministic finite element analysis;

[0045] In this embodiment, high stress level low cycle load and high frequency low amplitude high cycle load are considered at the same time to perform fluid-solid coupling analysis of turbine blades. Assuming that the high temperature airflow is turbulent and incompressible, fluid analysis is performed based on the standard ke turbulence model and the finite volume method, and thermal analysis is performed according to the temperature consistency principle to obtain the surface pressure distribution, velocity distribution and temperature distribution of the turbine blades; the surface pressure and temperature are introduced into the structural field analysis, fixed constraints are set and speed loads are loaded to obtain the stress and strain distribution of the turbine blades. The analysis results show that the stress level at the root of the turbine blade is relatively high, which is a weak point prone to high and low cycle composite fatigue failure.

[0046] Step 5, using collaborative sampling technology to extract a small batch of input samples of about 30-200, combined with deterministic finite element simulation results, to obtain a one-to-one corresponding output response of the dangerous parts of the turbine blade root, and establish a database of input parameters and response parameters containing correlation relationships;

[0047] In this embodiment, the collaborative sampling technology adopts the Latin hypercube sampling method to sample all input parameters of the turbine blades and the response parameters of each subject simultaneously for multiple times.

[0048] In this embodiment, 60 groups of input variables are extracted in a coordinated manner using collaborative sampling technology, and combined with the flow-solid deterministic finite element simulation of the turbine blade, 60 groups of output responses such as average stress, strain amplitude and stress amplitude corresponding to the blade root are obtained. The corresponding output responses such as low-cycle fatigue life and high-cycle fatigue life are obtained using each life model, and a sample point set corresponding to the input random variables and the responses of each subject is established.

[0049] Step 6: Establish different types of distributed agent models for each subject response, fit the agent model according to the input parameters and response parameter database established in step 5, and select the agent model with the highest accuracy as the optimal agent model for each subject response;

[0050] In this embodiment, when selecting the proxy model, according to the characteristics of the reliability analysis of each subject response, a proxy model with high accuracy is selected by minimizing the loss function. The proxy model types include polynomial model, Kriging model, support vector machine model, machine learning model and deep learning model.

[0051] In this embodiment, the sample point set is divided into training samples, verification samples and test samples in a ratio of 6:2:2, and the distributed agent model of high-cycle fatigue and low-cycle fatigue is established using the proposed distributed agent model.

[0052] Step 7, based on the resistance probability distribution model and the generalized "stress-resistance" interference method, establish a reliability model related to high-cycle fatigue and low-cycle fatigue failure;

[0053] In this embodiment, the generalized stress and resistance probability distribution model of the structure is obtained by finite element simulation to calculate the generalized strength of the structure, and the generalized "stress-resistance" interference method is used to establish a reliability model related to high-cycle fatigue and low-cycle fatigue failure.

[0054] Step 8: Use collaborative sampling technology to extract responses from various disciplines and use the optimal proxy model to complete the reliability analysis of turbine blades;

[0055] Step 8 includes the following sub-steps:

[0056] Step 81, using the optimal proxy model established by the responses of each subject to replace the real structure of the turbine blade for reliability analysis;

[0057] Step 82, based on the allowable value of each subject response, that is, judging whether the response of each subject after the turbine blade is loaded is too large or too small, a measurement standard is predetermined, and a multi-mode limit state equation is established. It is assumed that the turbine blade reliability analysis involves l response subjects, and the pth allowable output response is [Y (p) ], then the limit state equation under the p-th subject response is g (p) (x) can be expressed as:

[0058] (5)

[0059] in, is the optimal solution obtained by the p-th distributed agent model;

[0060] Step 83, using collaborative sampling technology to perform large-scale sampling of about 10,000-100,000 input parameters, and bringing them into the optimal proxy model to calculate the response parameters corresponding to high-cycle fatigue and low-cycle fatigue;

[0061] Step 84, judging whether the response value of each subject falls within the safety domain S according to the limit state equation in step 82, counting the number of sample points falling within the safety domain, and using the frequency of occurrence of the sample points within the safety domain to approximately replace the reliability value, that is, the reliability of the turbine blade is obtained, and the reliability analysis of the multi-failure mode of the turbine blade is reliably completed. The reliability calculation formula is as follows:

[0062] (6)

[0063] In the formula, R S is reliability; S (g) is the safety domain indicator function; M is the total number of sampling points; S is the safety domain; M S is the number of sample points in the security domain.

[0064] In this embodiment, the output responses of each distributed response surface are extracted in a coordinated manner using a collaborative sampling technique. A turbine blade multi-failure mode reliability model is established based on the established proxy model, and the low-cycle fatigue probability life and high-cycle fatigue probability life of the turbine blade are calculated.

[0065] Step 9: Import the reliability analysis results of the turbine blades into the Copula correlation model to evaluate the correlation between high-cycle fatigue and low-cycle fatigue using simulation results.

[0066] Step 9 includes the following sub-steps:

[0067] Step 91, based on the database of input parameters such as temperature, rotation speed, elastic modulus and response parameters such as average stress, strain amplitude and stress amplitude containing correlation established in step 5, solve the Copula parameters using the maximum likelihood estimation method;

[0068] Step 92, using AIC and BIC criteria to select the optimal Copula function type, and establish a Copula correlation model for correlation quantification;

[0069] For the Copula function C[u, v], u and v are marginal distribution functions, then the calculation formulas for the AIC and BIC index values ​​are as follows:

[0070] (7)

[0071] Among them, k is the number of marginal distribution functions, L(·) is the likelihood function, m is the sample size, and ρ is the correlation coefficient between the input variables. The AIC and BIC values ​​calculated by the Gaussian Copula function are the smallest, so the Gaussian Copula function is selected to establish the Copula correlation model for quantifying the correlation relationship.

[0072] Step 93, importing the reliability analysis result of the turbine blade calculated in step 8 into the Gaussian Copula correlation model, and obtaining an obvious lower tail correlation relationship between high cycle fatigue and low cycle fatigue.

[0073] The above embodiments are provided only for the purpose of describing the present invention, rather than for limiting the scope of the present invention. For those skilled in the art, the present invention may have various changes and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.

Claims

1. A distributed collaborative reliability method for multi-failure structures based on subject decomposition, characterized in that: The following steps are involved: Step S1, for a multi-failure structure including multiple failure modes, counting the types and quantities of failure modes of the multi-failure structure according to actual engineering practice; The multi-failure structure containing multiple failure modes refers to a complex engineering structure that produces multiple failure modes under the action of fluid load, thermal load, and solid load. The multiple failure modes include: low-cycle fatigue failure, high-cycle fatigue failure, and creep-fatigue failure; Step S2, disassembling the multiple failure modes according to the different disciplines to which they belong, and obtaining the output responses of the single disciplines, i.e., the responses of each discipline; the disciplines include fluid mechanics, thermodynamics, and solid mechanics; Step S3, according to the characteristics of each failure mode, respectively obtain the input random variable and output response of each failure mode, and use the input random variable and output response of each failure mode as the input parameter and response parameter of the multi-failure structure; the characteristics of each failure mode include that low-cycle fatigue failure requires a quasi-static response analysis for the state cycle, and high-cycle fatigue requires a dynamic response analysis for the structure; Step S4, performing deterministic analysis on the multi-failure structure, obtaining dangerous parts of the multi-failure structure, completing deterministic finite element analysis, and obtaining deterministic finite element simulation results; Step S5, using collaborative sampling technology to extract 30-200 small batch input samples, combined with deterministic finite element simulation results, to obtain one-to-one corresponding output responses of dangerous parts of the multi-failure structure, and establish a database of input parameters and response parameters containing correlations; Step S6, establishing different types of distributed agent models for each subject response, fitting the agent model according to the input parameters and response parameter database established in step S5, and selecting the agent model with the highest accuracy as the optimal agent model for each subject response; Step S7, establishing a reliability model related to multi-mode failure based on the resistance probability distribution model and the generalized "stress-strength" interference method; Step S8, based on the reliability model in step S7, using collaborative sampling technology to extract responses from various disciplines, using the optimal proxy model in step S6 to complete the reliability analysis of the multi-failure structure, and obtaining the reliability of the multi-failure structure; Step S9, importing the reliability result of the multi-failure structure obtained in step S8 into the Copula correlation model, quantitatively quantifying the correlation between the failure modes with the simulation results, and obtaining the correlation between the failure modes of the multi-failure structure. Only on the basis of the quantitative evaluation of the failure correlation can an accurate structural system reliability model be established, that is, the correlation between the failure modes of the multi-failure structure is obtained, and the reliability of the multi-failure structure is accurately predicted.

2. The multi-failure structure distributed collaborative reliability method based on subject decomposition according to claim 1 is characterized by: In step S5, the specific steps of the collaborative sampling technology are: using the Latin hypercube sampling method to sample all input parameters of the multi-failure structure and the response parameters of each subject simultaneously for multiple times.

3. The multi-failure structure distributed collaborative reliability method based on subject decomposition according to claim 1 is characterized by: In step S6, when selecting the proxy model with the highest accuracy as the optimal proxy model, according to the characteristics of the reliability analysis of the responses of each subject, the proxy model with high accuracy is selected as the optimal proxy model by minimizing the loss function; the types of the proxy models include polynomial models, kriging models, support vector machine models, machine learning models and deep learning models.

4. The multi-failure structure distributed collaborative reliability method based on subject decomposition according to claim 1 is characterized by: The step S8 comprises the following sub-steps: Step S81, using the optimal proxy model established by each subject response to replace the multi-failure structure for reliability analysis; Step S82, based on the allowable value of each subject response, that is, judging whether the response of each subject after the multi-failure structure is loaded is too large or too small, a standard of measurement is predetermined, and a multi-mode limit state equation is established. Assume that the reliability analysis of the multi-failure structure involves l response subjects, and the pth allowable output response is [Y (p) ], then the limit state equation under the p-th subject response is g (p) (x) is expressed as: (1) in, is the optimal solution obtained by the p-th distributed agent model; Step S83, using collaborative sampling technology to perform large-scale sampling of 10,000-100,000 on the input parameters, and bringing them into the optimal proxy model to calculate the response parameters corresponding to each failure mode; Step S84, judging whether the response value of each subject falls within the safety domain S according to the limit state equation in step S82, counting the number of sample points falling within the safety domain, and replacing the reliability value with the frequency of occurrence of the sample points within the safety domain, that is, obtaining the reliability of the multi-failure structure, completing the reliability analysis of the multi-failure structure, and the reliability calculation formula is as follows: (2) In the formula, R S is reliability; S (g) is the safety region indicator function; g represents the limit state function; M is the total number of sampling points; S is the safety zone; M S is the number of sample points in the security domain.

5. The multi-failure structure distributed collaborative reliability method based on subject decomposition according to claim 1 is characterized by: The step S9 comprises the following steps: Step S91, based on the database of input parameters and response parameters containing correlation established in step S5, solve the Copula parameters using the maximum likelihood estimation method; Step S92, using AIC and BIC criteria to select the Copula function type and establish a Copula correlation model for correlation quantification; Step S93, importing the reliability analysis result of the multi-failure structure calculated in step S8 into the Copula correlation model to evaluate the correlation between the multi-failure modes.

Images