Multilevel Reliability Evaluation Method for Aero-Engines Based on Unit Data Reorganization

Abstract

The present invention's multi-level reliability assessment method for aircraft engines based on unit data reorganization includes: clarifying the hierarchical and logical relationships between the engine system and its subsystems, components, and parts based on the engine structure and fault tree; randomly sampling and combining the units according to the hierarchical relationships to form the engine system using the Monte Carlo method; and deriving equivalent life data samples or mixed life data samples for the entire system based on the life data of each unit according to the logical relationship; and using the equivalent life data samples or mixed life data samples of the entire system to estimate the system life probability distribution parameters using existing methods, thereby achieving a direct assessment of system reliability from unit life samples. The present assessment method is applicable to various life distribution types and can more accurately, robustly, and effectively assess aircraft engine reliability using multi-level data while maintaining sufficient conservatism.

Description

Technical Field

[0001] The invention belongs to the technical field of reliability assessment of aircraft engine systems and relates to a multi-level reliability assessment method for aircraft engines based on unit data reorganization. Background Art

[0002] Aircraft engines are expensive, making system-level reliability testing prohibitively expensive and difficult to conduct. Consequently, a "pyramid-style" testing approach is typically employed, involving numerous component tests, a smaller number of subassembly and subsystem tests, and a very small number of complete aircraft tests. However, fully utilizing this multi-layered data set to accurately assess system reliability remains a complex and unresolved issue.

[0003] Traditional methods for complex system reliability synthesis mainly include confidence interval estimation and Bayesian methods. Approximate confidence limit methods, including the LM method, MML method, SR method, and CMSR method, are widely used. Regarding the distribution of lifespan, traditional methods are mostly based on binomial and exponential distributions. Research at home and abroad has shown that the three-parameter Weibull distribution has strong fitting capabilities and is more realistic, leading to its increasing application in product lifespan and reliability assessment. Regarding system reliability models, existing methods for series systems like aircraft engines use a series system reliability product model, where system reliability is equal to the product of the reliabilities of each unit. This model has a significant drawback: unit reliability errors are amplified in the system reliability assessment results in the form of a product. For complex mechanical systems with high reliability and long lifespans like aircraft engines, test data for subsystems and components is extremely limited. Small sample sizes lead to large errors in unit reliability calculations. After multi-level synthesis, these unit errors accumulate in the system, causing the assessment results to deviate significantly from reality. Therefore, a more accurate, robust, and effective multi-level reliability assessment method for aircraft engines is urgently needed. Summary of the Invention

[0004] Traditional complex system reliability synthesis methods are limited to binomial distribution and exponential distribution. The use of a series system reliability product model has the disadvantages of amplifying unit errors in a multiplicative manner and having poor evaluation accuracy when the sample size is small. The present invention provides an aircraft engine multi-level reliability assessment method based on unit data reorganization. This method is applicable to various life distribution types and produces more accurate and robust results.

[0005] The multi-level reliability assessment method for an aero-engine based on unit data reorganization of the present invention comprises:

[0006] Step 1: Based on the engine structure and fault tree, define the system boundary of the engine reliability assessment and clarify the hierarchical and logical relationships between the engine system and its subsystems, components, and parts.

[0007] Step 2: Using the Monte Carlo method, randomly sample and combine the subsystems, components, and parts used to obtain life data in the test according to the hierarchical relationship to form the engine system. According to the logical relationship, the life data of each unit in the system is used to derive the life data of the engine system, forming an equivalent life data sample or a mixed life data sample of the whole system;

[0008] Step 3: Using the equivalent life data samples or mixed life data samples of the whole system in step 2, estimate the system life probability distribution parameters through existing methods according to the data distribution type, and realize direct evaluation from unit life samples to system reliability.

[0009] The multi-level reliability assessment method for aircraft engines based on unit data reorganization of the present invention has at least the following beneficial effects:

[0010] 1) The present invention avoids the propagation and amplification of unit errors in the process of step-by-step reliability synthesis and conversion in traditional methods, and can use multi-level data to evaluate aircraft engine reliability more accurately, robustly and effectively under the premise of being sufficiently conservative.

[0011] 2) The present invention has strong versatility and is applicable to various types of life data. Ideal results can be obtained for binomial distribution, exponential distribution, lognormal distribution, and two-parameter and three-parameter Weibull distributions. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of the multi-level reliability assessment method of aircraft engines based on unit data reorganization. DETAILED DESCRIPTION

[0013] like Figure 1 As shown, the aircraft engine multi-level reliability assessment method based on unit data reorganization of the present invention includes the following steps:

[0014] Step 1: Based on the engine structure and fault tree, define the system boundary of the engine reliability assessment, and clarify the hierarchical and logical relationships between the engine system and subsystems, components, and parts. Specifically:

[0015] 1) Analyze components from the perspective of failure physics based on the aircraft engine's service load environment, component functions, materials, and design criteria. Components that have a significant impact on overall system reliability and require consideration, along with their corresponding components and subsystems, are selected, and system boundaries are defined.

[0016] 2) The units that affect the reliability of the entire system are divided into three levels: subsystems, components, and parts. A subsystem consists of several components, and a component is composed of several parts.

[0017] 3) Draw a pyramid reliability diagram to analyze and express the hierarchical and logical relationships between the entire system and its constituent units.

[0018] Step 2: Using the Monte Carlo method, randomly sample and combine the subsystems, components, and parts used to obtain life data in the test according to the hierarchical relationship to form the engine system. According to the logical relationship, the life data of each unit in the system is used to obtain the life data of the engine system, forming an equivalent life data sample or a mixed life data sample of the whole system. Specifically:

[0019] Step 2.1: Based on the engine hierarchy, starting from the lowest level, statistically reconstruct the component-level unit life data, transfer the life information to the corresponding components, and obtain equivalent component life data samples or mixed component life data samples. The specific process is as follows:

[0020] 1) Data collation: Ensure that all components of the subassembly have lifespan data and that the data is reliable. The sample sizes of each component should not vary significantly, and samples that are too small should be supplemented. For components with identical components, estimate the lifespan distribution parameters based on the component life test data, and then perform random sampling using the Monte Carlo method. Each set of values drawn is equivalent to obtaining a lifespan sample of the same unit, which is used for sample reconstruction.

[0021] 2) Using the Monte Carlo method, randomly select one component from each sample of the component to form a component sample;

[0022] 3) Based on the logical relationships between components and subassemblies, the life data of the subassembly sample is obtained. In aircraft engines, most units are connected in series. The random variable of the life of a series system is equal to the minimum order statistic of the random variables of the life of all units that make up the system. Therefore, the life of the subassembly sample is the minimum value of its component life. For a parallel system, the system life is the maximum order statistic of the unit life. In this case, the maximum value of the component life is taken as the subassembly life.

[0023] 4) According to 2) and 3), sampling with replacement n times can obtain n component samples, and obtain a set of equivalent life data samples of components obtained by recombining the life data of components. Obviously, the number of components n that can be combined depends on the number of parts with the smallest sample size.

[0024] 5) If there is real life data of components, then the equivalent life data samples of the components and the real life data of the components are combined to obtain the mixed life data samples of the components; when the test conditions of the components and parts are consistent, the two types of data are directly mixed; in practice, the test loads and test environments of components and parts are sometimes different. In this case, if the environmental factors are known, the equivalent life data samples of the components are corrected by the environmental factors; if the environmental factors are unknown, the dispersion of the system life data should be within a certain range, and the equivalent life data samples of the components are corrected according to the 6σ principle, that is, if the maximum distance between the real life data and the equivalent life data samples of the components is outside 6 times the standard deviation of the latter, then the equivalent life data samples of all components are shifted to ensure that the maximum distance is 6σ.

[0025] Step 2.2: Integrate the life data step by step upward according to step 2.1 until an equivalent life data sample or a mixed life data sample of the entire system is obtained.

[0026] For example, consider a series system consisting of 10 success / failure components, each with a reliability of 0.99. Each component undergoes 30 tests, with the number of failures being 0, 0, 0, 1, 0, 1, 0, 0, 0, 0. Assume that one component is randomly sampled from each of the 30 samples. These 10 components constitute a system sample. If any component in the system fails, the system fails. If 30 system samples are randomly sampled with replacement, with a number of failures of 2, the corresponding system reliability can be calculated.

[0027] The method of this invention can also be conveniently applied to situations where component lifespans follow exponential, lognormal, two-parameter, or three-parameter Weibull distributions. For example, a rotor system consisting of a disk, blades, and a pair of bearings has a three-parameter Weibull distribution with parameters W(2.5, 1600, 1200), W(2.0, 2500, 1250), and W(1.5, 4000, 1400). The following component test data is obtained by simulation:

[0028] Roulette: 2916.81, 3556.17, 2469.61, 1327.78, 2000.86, 2543.84, 1780.31, 2133.04, 2256.64, 2676.14;

[0029] Blade: 4173.82, 3599.62, 5334.78, 5273.03, 3964.35, 2272.88, 2807.52, 4361.39, 3388.10, 2799.64, 3705.09, 2709.31, 1553.68, 3277.68, 2318.39;

[0030] Bearings:3851.66,2950.24,1924.31,6167.94,3464.53,5042.58,5437.47,1500.82,4177.06,2271.23.

[0031] The correlation coefficient method is used to fit the bearing distribution parameters W (1.79, 3420.71, 699.77). Based on this, another set of bearing life data is generated by simulation: 2746.59, 5875.28, 1921.16, 2138.93, 4564.06, 1959.69, 3199.40, 4355.54, 7427.48, 5937.16.

[0032] A data is randomly selected from each of the four components. The minimum life span is the system life span. The minimum sample size of the four components is 10. Ten random samples are drawn with replacement to obtain a set of system "equivalent life span samples": 1780.31, 1500.82, 2256.64, 2133.04, 2256.64, 1921.16, 2256.64, 1780.31, 2271.23, 1780.31.

[0033] Step 3: Using the equivalent life data samples or mixed life data samples of the entire system in step 2, estimate the system life probability distribution parameters using existing methods according to the data distribution type, and achieve direct evaluation from unit life samples to system reliability to avoid error propagation. Specifically:

[0034] 1) Determine the distribution of the whole machine life data based on experience. Common distribution types include binomial distribution, exponential distribution, normal distribution, lognormal distribution, and two-parameter and three-parameter Weibull distribution. Different distribution types have corresponding distribution test methods to verify whether the selected distribution type is appropriate;

[0035] 2) According to the life data distribution type, the existing relevant methods are used to estimate the distribution parameters using the equivalent life data samples or mixed life data samples of the whole system, and then estimate the reliability R of the whole system;

[0036] 3) Using the Bootstrap method, repeat the above steps multiple times to obtain multiple reliability values of the entire system R. Based on this, the reliability of the entire system is estimated, and the reliability confidence lower limit R under the confidence level γ is calculated by the following formula: L :

[0037] For the series system consisting of 10 success / failure components in step 2, repeating the above process multiple times, such as 10,000 times, we can calculate that the mean of the system reliability is 0.9344 and the standard deviation is 0.0448. From this, we can obtain the lower confidence limit of the system reliability. The reliability of the system at 95% confidence level is 0.8607.

[0038] If the traditional LM method is used, the point estimate of system reliability is 0.9344, and the reliability at 95% confidence level is 0.8062.

[0039] The precise value of the system reliability is easily calculated to be 0.9044. Compared with the LM method, the method provided by the present invention provides consistent point estimates and a more accurate lower confidence limit for reliability at a 95% confidence level. Multiple simulations of the aforementioned success-failure system with a sample size of 30 were conducted. The lower confidence limits for the system reliability calculated by this method at a 95% confidence level ranged from 0.63 to 0.90, both exceeding those estimated by the LM method, demonstrating the method's sufficient conservatism and effectiveness.

[0040] For the rotor system example in Step 2, the reliability lower confidence limit at a 95% confidence level after 1500 hours of operation was evaluated. Based on the aforementioned "equivalent life sample" of the system, the correlation coefficient method was used to obtain the distribution parameters W(4.78, 1333.16, 768.05), and the reliability after 1500 hours of operation was calculated to be 0.9445. Repeating the "system test" multiple times to obtain 10,000 "equivalent life samples" of the system, the reliability after 1500 hours of operation had a mean of 0.86, a standard deviation of 0.085, and a reliability lower confidence limit of 0.7197.

[0041] Using traditional methods to fit the distribution parameters, the reliability of the disk, blade, and bearing calculated after 1500 hours is 0.9019, 0.9652, and 0.9284, respectively. Using the series system model, the point estimate of the system reliability is 0.7503. Further calculations show that the equivalent number of system tests is 10, the equivalent number of failures is 2.4968, and the lower confidence limit of the reliability at a 95% confidence level is 0.3444.

[0042] Based on the true distribution of each component, the exact solution to the system reliability is easily calculated to be 0.9675. This demonstrates that the method provided by the present invention significantly outperforms traditional methods for small sample sizes. For larger samples, such as 30, the reliability mean obtained by this method is 0.9174, the standard deviation is 0.038, and the lower confidence limit is 0.8547. The traditional method achieves a reliability point estimate of 0.9103 and a lower confidence limit of 0.7340, demonstrating a clear advantage for this method as well.

Claims

1. A multi-level reliability assessment method for aircraft engines based on unit data reorganization, characterized in that: include: Step 1: Based on the engine structure and fault tree, define the system boundary of the engine reliability assessment and clarify the hierarchical and logical relationships between the engine system and its subsystems, components, and parts. Step 2: Using the Monte Carlo method, randomly sample and combine the subsystems, components, and parts used to obtain life data in the test according to the hierarchical relationship to form the engine system. According to the logical relationship, the life data of each unit in the system is used to derive the life data of the engine system, forming an equivalent life data sample or a mixed life data sample of the whole system; Step 3: Using the equivalent life data samples or mixed life data samples of the whole system in step 2, estimate the system life probability distribution parameters through existing methods according to the data distribution type, and realize direct evaluation from unit life samples to system reliability.

2. The method for multi-level reliability assessment of an aircraft engine based on unit data reorganization according to claim 1, characterized in that: In step 1, the hierarchical and logical relationships between the engine and its subsystems, components, and parts are clarified as follows: 1) Analyze components from the perspective of failure physics based on the aircraft engine's service load environment, component functions, materials, and design criteria. Components that have a significant impact on overall system reliability and require consideration, along with their corresponding components and subsystems, are selected, and system boundaries are defined. 2) The units that affect the reliability of the entire system are divided into three levels: subsystems, components, and parts. A subsystem consists of several components, and a component is composed of several parts. 3) Draw a pyramid reliability diagram to analyze and express the hierarchical and logical relationships between the entire system and its constituent units.

3. The method for multi-level reliability assessment of an aircraft engine based on unit data reorganization according to claim 1, characterized in that: The equivalent system life test samples formed in step 2 are specifically: Step 2.1: Based on the engine hierarchy, starting from the lowest level, statistically reconstruct the component-level unit life data, transfer the life information to the corresponding components, and obtain equivalent component life data samples or mixed component life data samples; Step 2.2: Integrate the life data step by step upward according to step 2.1 until an equivalent life data sample or a mixed life data sample of the entire system is obtained.

4. The method for multi-level reliability assessment of an aircraft engine based on unit data reorganization according to claim 3, characterized in that: The step 2.1 is specifically as follows: 1) Data collation: Ensure that all components of the subassembly have lifespan data and that the data is reliable. The sample sizes of each component should not vary significantly, and samples that are too small should be supplemented. For components with identical components, estimate the lifespan distribution parameters based on the component life test data, and then perform random sampling using the Monte Carlo method. Each set of values drawn is equivalent to obtaining a lifespan sample of the same unit, which is used for sample reconstruction. 2) Using the Monte Carlo method, randomly select one component from each sample of the component to form a component sample; 3) Based on the logical relationships between components and subassemblies, the life data of the subassembly sample is obtained. In aircraft engines, most units are connected in series. The random variable of the life of a series system is equal to the minimum order statistic of the random variables of the life of all units that make up the system. Therefore, the life of the subassembly sample is the minimum value of its component life. For a parallel system, the system life is the maximum order statistic of the unit life. In this case, the maximum value of the component life is taken as the subassembly life. 4) Sampling with replacement n times according to 2) and 3) can obtain n component samples, and obtain a set of equivalent component life data samples obtained by recombining the component life data. Obviously, the number of components n that can be combined depends on the number of parts with the smallest sample size. 5) If there is real life data of components, then the equivalent life data samples of the components and the real life data of the components are combined to obtain the mixed life data samples of the components; when the test conditions of the components and parts are consistent, the two types of data are directly mixed; in practice, the test loads and test environments of components and parts are sometimes different. In this case, if the environmental factors are known, the equivalent life data samples of the components are corrected by the environmental factors; if the environmental factors are unknown, the dispersion of the system life data should be within a certain range, and the equivalent life data samples of the components are corrected according to the 6σ principle, that is, if the maximum distance between the real life data and the equivalent life data samples of the components is outside 6 times the standard deviation of the latter, then the equivalent life data samples of all components are shifted to ensure that the maximum distance is 6σ.

5. The method for multi-level reliability assessment of an aircraft engine based on unit data reorganization according to claim 1, characterized in that: The step 3 is specifically as follows: 1) Determine the distribution of the whole machine life data based on experience. Different distribution types have corresponding distribution test methods to verify whether the selected distribution type is appropriate; 2) Based on the life data distribution type, the existing method is used to estimate the distribution parameters using the equivalent life data samples or mixed life data samples of the whole system, and then the reliability R of the whole system is estimated; 3) Using the Bootstrap method, repeat the above steps multiple times to obtain multiple reliability values of the entire system R. Based on this, the reliability of the entire system is estimated, and the reliability confidence lower limit R under the confidence level γ is calculated by the following formula: L : In the formula s R are the mean and standard deviation of the reliability R of the whole system, μ 1-γ is the 1-γ quantile of the standard normal distribution.

Images