Imprecise probability input reliability sensitivity analysis method based on adaptive kriging model
Through the combination of the adaptive Kriging model and the sampling probability density function, the problem of time-consuming and costly calculation of reliability sensitivity analysis under parameter uncertainty is solved, and efficient non-accurate probability input reliability sensitivity analysis is achieved, suitable for complex engineering question.
Patent Information
- Application Number
- CN202410352369.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-26
- Publication Date
- 2024-11-22
AI Technical Summary
In the prior art, reliability sensitivity analysis under parameter uncertainty requires repeated calls to function functions, resulting in increased calculation time and cost, which makes it difficult to effectively apply in engineering.
Using the inaccurate probability input reliability sensitivity analysis method based on the adaptive Kriging model, the initial training sample set is extracted from the distribution parameters of the target structure by sampling probability density function, the initial Kriging model is constructed, and adaptive learning is performed through the U learning function. Until converge, the failure probability and its coefficient of variation are solved, the sample set is adjusted to meet the requirements of the coefficient of variation, and the main sensitivity and total sensitivity index are finally calculated.
It significantly improves the computing efficiency, reduces the sample pool size, reduces the calculation cost, and can quickly and efficiently perform reliability sensitivity analysis under inaccurate probability input, suitable for complex limit state functions and implicit functional functions.
Smart Images

Figure CN119004937A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of deterministic analysis and relates to an inexact probability input reliability sensitivity analysis method based on an adaptive Kriging model. Background Art
[0002] In practical engineering, such as aerospace engineering and civil engineering, the existence of uncertainty is inevitable. Generally speaking, we can classify uncertainty into two types, random uncertainty caused by natural changes and epistemic uncertainty caused by lack of knowledge. Random uncertainty arises from the inherent variability of the system, while epistemic uncertainty arises from the lack of knowledge of basic phenomena and can be reduced with the increase of knowledge or the increase of observation data. Usually, random uncertainty will lead to accidental failure or lack of robustness of engineering structures, while epistemic uncertainty will hinder our understanding of structural reliability and dependability. In order to reduce the impact of uncertainty on structural systems, uncertainty analysis has been widely used to help researchers evaluate the failure risk of model structures, including forward analysis and reverse analysis. Forward analysis mainly studies how the uncertainty of model input affects the model output; reverse analysis, namely sensitivity analysis, mainly studies how the uncertainty of model output is distributed to the model input uncertainty from different sources, so as to determine which input parameters are more important to the output uncertainty. As an important research content of sensitivity analysis, reliability sensitivity analysis is a key part of reliability design. Reliability sensitivity analysis can accurately measure the importance of the uncertainty of different input variables to the failure probability by analyzing the influence of each input variable alone or the interaction between variables on the failure probability. Sort by importance measure, ignore variables with smaller impact, find out variables with greater impact on structural failure probability for design optimization, and effectively improve system safety performance.
[0003] At present, many methods have been developed for reliability sensitivity analysis under single random uncertainty input. However, there is no accurate and reliable calculation method to study and optimize the importance measurement index of the widely existing non-precise probability input in engineering practice, that is, the distribution parameters of random input variables also have cognitive uncertainty. Therefore, it is very necessary to study the reliability sensitivity analysis method under this situation, which can provide important guidance for the reliability design and optimization of structural systems with distributed parameter uncertainty input in practice.
[0004] Chen Zhiyuan et al. proposed a global sensitivity index based on reliability for distribution parameters in their article "An Efficient Reliability Sensitivity Analysis Method under Parameter Uncertainty", and established an efficient method for solving the reliability sensitivity index of distribution parameters by single-layer sampling by combining the Surrogate Sampling Probability Density Function (SSPDF) with the Importance Sampling (IS) and Truncated Importance Sampling (TIS). However, in engineering practice, repeatedly calling the function will increase the computational time and cost, which is unacceptable for engineering problems. Summary of the invention
[0005] The purpose of the present invention is to solve the technical problem that the efficient reliability sensitivity analysis method under parameter uncertainty in the prior art requires repeated calls to functional functions, which increases the calculation time and cost, and to provide an imprecise probability input reliability sensitivity analysis method based on an adaptive Kriging model.
[0006] In order to achieve the above object, the present invention adopts the following technical solutions:
[0007] In a first aspect, the present invention discloses a reliability sensitivity analysis method for inexact probability input based on an adaptive Kriging model, comprising the following steps:
[0008] Extracting an initial training sample set from the distribution parameters of the target structure based on a sampling probability density function, and constructing an initial Kriging model based on the initial training sample set;
[0009] Adaptively learning the initial Kriging model based on the U learning function; when the adaptive learning process converges, solving the failure probability and its coefficient of variation, otherwise, reconstructing the initial Kriging model based on the initial training sample set until its adaptive learning process converges;
[0010] When the coefficient of variation meets the requirement, an adaptive Kriging model is obtained; otherwise, the initial training sample set is expanded, and the initial Kriging model is reconstructed based on the expanded initial training sample set until its coefficient of variation meets the requirement;
[0011] Generate a sample matrix of input variables based on a sampling probability density function, substitute samples in the sample matrix into the adaptive Kriging model, and obtain corresponding output values and their indicator functions;
[0012] The main sensitivity index and the total sensitivity index of the target structure are calculated based on the indicator function, and reliability sensitivity analysis is performed.
[0013] The present invention is further improved in that:
[0014] The step of extracting an initial training sample set from the distribution parameters of the target structure based on the sampling probability density function specifically includes:
[0015] Extract the sample matrix of the distribution parameters of the target structure;
[0016] Based on the sampling probability density function, several groups of samples of the input variables are extracted as the sample pool;
[0017] A number of input-output samples in the process of solving the design point based on the sample matrix and the sample pool are brought into the original function, and a number of groups of function values are obtained as initial training sample sets.
[0018] The adaptive learning of the initial Kriging model based on the U learning function and judging whether the adaptive learning process converges specifically include:
[0019] Calculate the U learning function value of each sample point in the sample pool based on the initial Kriging model;
[0020] Select and update the sample points according to the U learning function value of each sample point to obtain the U learning sample points;
[0021] When the U learning function value of the U learning sample point is ≥2, the adaptive learning process converges.
[0022] When the adaptive learning process does not converge, reconstructing the initial Kriging model includes:
[0023] The actual model output value of the U learning sample point is calculated, and the coordinate point composed of the U learning sample point and its actual model output value is added to the initial training sample set to reconstruct the initial Kriging model.
[0024] The solution of failure probability and its coefficient of variation specifically includes:
[0025] Predicting the failure domain indicator function of all samples in the sample pool based on the initial Kriging model, thereby obtaining a failure probability estimate as the failure probability;
[0026] Through several samplings, several failure probability estimates are obtained and their means and variances are calculated, and then the corresponding coefficient of variation is calculated.
[0027] When the coefficient of variation meets the requirement, an adaptive Kriging model is obtained; otherwise, the initial training sample set is expanded and the initial Kriging model is reconstructed, which specifically includes:
[0028] When the coefficient of variation is less than 5%, an adaptive Kriging model is obtained;
[0029] When the coefficient of variation is ≥5%, the initial training sample set is expanded and the initial Kriging model is reconstructed.
[0030] The sampling probability density function includes an importance sampling probability density function and a truncated importance sampling probability density function.
[0031] The generating of a sample matrix of input variables based on a sampling probability density function, substituting samples in the sample matrix into the adaptive Kriging model, and obtaining corresponding output values and their indicator functions specifically include:
[0032] When the sampling probability density function is an important sampling probability density function:
[0033] Generate independent N×n dimensional sample matrices A and B of the input variables based on the importance sampling probability density function:
[0034]
[0035]
[0036] Assign the elements of matrix A to C (i) , except for the i-th column element, we get:
[0037]
[0038] The matrices A, B and C (i) Substitute the samples in into the adaptive Kriging model and obtain the corresponding output value: Y kA =g k (A), Y kB =g k (B) Then determine whether the output value falls within the failure domain F = {x:g(x)≤0} and find the corresponding indicator function:
[0039] I FA =I F (A), I FB =I F (B)
[0040] The generating of a sample matrix of input variables based on a sampling probability density function, substituting samples in the sample matrix into the adaptive Kriging model, and obtaining corresponding output values and their indicator functions specifically include:
[0041] When the sampling probability density function is a truncated importance sampling probability density function, the following steps are also included:
[0042] The matrices A, B and C are transformed by β-sphere (i) The samples falling into the safe region are cut off, and the corresponding indicator function is obtained: I βA =I β (A), I βB =I β (B)
[0043] The sample points whose β-ball indicator function is not 0 are re-formed into a new matrix, denoted as A', B', C' (i) , and then calculate the value of the new matrix output function through the newly fitted adaptive Kriging model, that is, the corresponding input-output value: Y′ kA =g k (A'), Y' kB =g k (B'),
[0044] After obtaining the output values of the new matrix, estimate whether these sample points are failed and obtain the corresponding failure domain indicator function: I' FA =I F (A'),I' FB =I F (B'),
[0045] By discarding the sample points whose beta-sphere indicator function is 0, there is no need to calculate their performance function values again. Instead, they can be directly replaced with 0 at the original positions in the matrix of the failure domain indicator function, thereby forming a new matrix.
[0046] In a second aspect, the present invention discloses an inexact probability input reliability sensitivity analysis system based on an adaptive Kriging model, comprising:
[0047] An initial Kriging model building module extracts an initial training sample set from the distribution parameters of the target structure based on a sampling probability density function, and builds an initial Kriging model based on the initial training sample set;
[0048] An adaptive learning module performs adaptive learning on the initial Kriging model based on the U learning function; when the adaptive learning process converges, the failure probability and its coefficient of variation are solved; otherwise, the initial Kriging model is rebuilt based on the initial training sample set until the adaptive learning process converges;
[0049] An adaptive Kriging model construction module, when the coefficient of variation meets the requirements, an adaptive Kriging model is obtained; otherwise, the initial training sample set is expanded, and the initial Kriging model is reconstructed based on the expanded initial training sample set until its coefficient of variation meets the requirements;
[0050] An indicator function calculation module generates a sample matrix of input variables based on a sampling probability density function, substitutes samples in the sample matrix into the adaptive Kriging model, and obtains corresponding output values and their indicator functions;
[0051] The sensitivity analysis module calculates the main sensitivity index and the total sensitivity index of the target structure based on the indicator function, and performs reliability sensitivity analysis.
[0052] Compared with the prior art, the present invention has the following beneficial effects:
[0053] The present invention discloses a reliability sensitivity analysis method for inexact probability input based on an adaptive Kriging model. The adaptive Kriging (AK) model is combined with a sampling probability density function, and the implicit and complex limit functional state function in actual engineering is replaced by an explicit and simpler model, showing significant superiority in the global reliability sensitivity analysis of inexact probability input.
[0054] Furthermore, the sampling probability density function includes the importance sampling probability density function (SISPDF) and the truncated importance sampling probability density function (STISPDF). The adaptive Kriging model is combined with the importance sampling probability density function (SISPDF) and the truncated importance sampling probability density function (STISPDF). The efficient methods based on the adaptive Kriging model are proposed respectively: the single-layer parameter uncertainty reliability sensitivity analysis based on the adaptive Kriging model combined with SISPDF (AKSL-IS) method and the single-layer parameter uncertainty reliability sensitivity analysis based on the adaptive Kriging model combined with STISPDF (AKSL-TIS) method. The example results show that the AKSL-IS and AKSL-TIS methods update the important SS-PDF, update the sampling center through iterative design points, so that the sample points are as far as possible in the failure domain, reduce the sample pool size, and improve the calculation efficiency. The AKSL-TIS method introduces the β-sphere indicator function, and by screening the sample points that fall into the β-sphere, there is no need to calculate their model output values, so that the calculation efficiency is further improved. In addition, for extremely complex limit state functions or implicit functional functions, it takes a long time for MCS to call the model once. By embedding the adaptive Kriging proxy model to convert it into a simple explicit function, reliability analysis can be performed more quickly and efficiently. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0056] Figure 1 It is a flow chart of a reliability sensitivity analysis method of inexact probability input based on an adaptive Kriging model in the present invention;
[0057] Figure 2 A comparison diagram of the convergence trends of AKSL-IS and AKSL-TIS methods in an inexact probability input reliability sensitivity analysis method based on an adaptive Kriging model in the present invention; Figure 2 (a) is the convergence diagram of the coefficient of variation of the main sensitivity index of the most important variable h under the two methods as the sample pool size changes; Figure 2 (b) is the convergence diagram of the coefficient of variation of the total sensitivity index of the most important variable h under the two methods as the sample pool size changes;
[0058] Figure 3A schematic diagram of a headless rivet riveting process in an embodiment of an inexact probability input reliability sensitivity analysis method based on an adaptive Kriging model in the present invention;
[0059] Figure 4 This is a system diagram of reliability sensitivity analysis based on an adaptive Kriging model for inexact probability input in the present invention;
[0060] Figure 5 This is a module diagram of an electronic device in the present invention. DETAILED DESCRIPTION
[0061] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.
[0062] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0063] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.
[0064] The present invention is further described in detail below in conjunction with the accompanying drawings:
[0065] See also Figure 1 The embodiment of the present invention discloses a reliability sensitivity analysis method for inexact probability input based on an adaptive Kriging model, comprising the following steps:
[0066] S1, extracting an initial training sample set from the distribution parameters of the target structure based on a sampling probability density function, and constructing an initial Kriging model based on the initial training sample set;
[0067] S2, adaptively learning the initial Kriging model based on the U learning function; when the adaptive learning process converges, solving the failure probability and its coefficient of variation, otherwise, reconstructing the initial Kriging model based on the initial training sample set until its adaptive learning process converges;
[0068] S3, when the coefficient of variation meets the requirement, an adaptive Kriging model is obtained; otherwise, the initial training sample set is expanded, and the initial Kriging model is reconstructed based on the expanded initial training sample set until its coefficient of variation meets the requirement;
[0069] S4, generating a sample matrix of input variables based on the sampling probability density function, substituting the samples in the sample matrix into the adaptive Kriging model to obtain corresponding output values and their indicator functions;
[0070] S5, calculating the main sensitivity index and the total sensitivity index of the target structure based on the indicator function, and performing reliability sensitivity analysis.
[0071] The present invention discloses a reliability sensitivity analysis method for inaccurate probability input based on an adaptive Kriging model, which combines the adaptive Kriging model with a sampling probability density function, replaces the implicit and complex limit function state function in actual engineering with an explicit and simpler model, and shows significant superiority in the global reliability sensitivity analysis of inaccurate probability input. The relevant theories of the present invention are introduced as follows:
[0072] (1) Definition of global reliability sensitivity index
[0073] When only considering the uncertainty of input variables, in the reliability model, it is assumed that the performance function of the structure is Y = g(X1, X2, …, X n ), where g(X1,X2,…,X n ) is the performance response function, Y is the model output, X=(X1,X2,…,X n ) T It is an n-dimensional independent basic random variable with uncertainty. X (x) is expressed as the joint probability density function of n-dimensional independent random variables.
[0074] The limit state equation g(x)=0 divides the input variable space into two parts, corresponding to the failure domain F={x:g(x)≤0} and the safety domain F={x:g(x)>0}:
[0075] Therefore, the failure probability P f It can be expressed as: Where R n represents the n-dimensional sample space, and E(·) represents the expectation operator.
[0076] If the random variable X i Fixed at a given value When , we can obtain the conditional failure probability Right now The conditional failure probability calculation expression at this time can also be expressed through the failure domain indicator function as follows: Among them, R n-1 Represents n-1 dimensional input variable X ~i The integral space of Indicates removal of X i Then, the distribution functions of other variables.
[0077] When I F When X is used as output, the reliability sensitivity index based on the failure probability is equivalent to the indicator function I F Reliability sensitivity index based on variance. The main sensitivity index is: And the overall sensitivity index is: Where X ~i Contains except X i All input variables except
[0078] (2) Definition of global reliability sensitivity index under distributed parameter uncertainty
[0079] When there is an input variable X=(X1,X2,…X n ) T , and there is epistemic uncertainty in the distribution parameters of the input variables, we can see that in the entire calculation process there is both random uncertainty in X and Θ X The epistemic uncertainty of . Assuming the distribution parameters The dimension of is m, where is the input variable X i The corresponding m i dimensional distribution parameter, m1+m2+…+m n = m. We can describe Θ in many ways. X The epistemic uncertainty of the input parameter Θ is X The probability density function of To describe Θ X When the distribution parameters of X are uncertain due to epistemic uncertainty, the sensitivity index value based on reliability also changes with the distribution parameters Θ. X , and is no longer a fixed value.
[0080] In order to better measure the impact of input variable X on structural uncertainty under parameter uncertainty, it is necessary to consider it from a global perspective and analyze the average impact of input variables on failure probability in the distribution space. Therefore, the reliability sensitivity index of input variables under distribution parameter uncertainty can be expressed by the following formula:
[0081]
[0082]
[0083] in, represents the unconditional variance.
[0084] Main importance measure S i It reflects the single variable X i Output P for the structure in the distribution space f The mean of the impact, but the overall sensitivity It represents not only a single input variable, but also the interaction between input variables X on the structure output P in the entire distribution space. f The mean of the impact.
[0085] (3) Proxy Sampling Probability Density Function
[0086] In the presence of parameter uncertainty, the epistemic uncertainty of the distribution parameters is also expressed in probability theory, assuming that the distribution type and other model parameters are known. By introducing the "surrogate sampling probability density function (SSPDF)" in the reliability-based sensitivity analysis, the three-layer nested Monte Carlo calculation is simplified to a two-layer calculation.
[0087] The sample points extracted by a suitable SSPDF should cover the entire possible value area of the original input as much as possible. For the situation where the distribution parameters are uncertain and changing, in order to obtain a good distribution of the input variables, it is necessary to select according to the actual situation, so as to further obtain a suitable SSPDF. The existing method has clearly given the selection method of SSPDF. When performing global reliability sensitivity analysis on the small failure probability problem in engineering, a large amount of model operations are still required, which is difficult to be accepted by engineering practice. The present invention will adopt more efficient importance sampling method (IS) and truncated importance sampling method (TIS) to process the global reliability sensitivity analysis problem of small failure probability under parameter uncertainty in engineering.
[0088] The basic idea of the IS method is to introduce an important sampling density function and use it to replace the original probability density function for sampling. The extracted important sampling samples contribute more to the calculation of failure probability than the samples of the original distribution. The importance sampling method is more efficient in solving small failure probability problems because it can ensure that there are enough sample points falling within the failure domain. A new proxy important sampling probability density function is constructed by iterating the sampling center of the proxy sampling density function at the design point.
[0089] Example 1 AKSL-IS method for calculating input variable reliability sensitivity index
[0090] For the problem of small failure probability in engineering practice, Monte Carlo Simulation (MCS) usually extracts a large number of samples to ensure that there are enough sample points falling within the failure domain. However, a large sample pool will cause each Kriging adaptive learning to take a long time, increasing the computational cost. In order to solve the problem of excessive sample pool capacity under small failure probability, the present invention combines the proxy important sampling probability density function with the adaptive Kriging (AK) model, and proposes a single-layer parameter uncertainty reliability sensitivity analysis (AKSL-IS) method based on adaptive Kriging combined with SISPDF.
[0091] The proxy importance sampling probability density function is obtained by the method in the article "Efficient reliability sensitivity analysis method under parameter uncertainty" Afterwards, use Generate a sample pool, and use the sample points used in the iterative design point as the initial training set to build the initial Kriging model. Then use the U learning function to adaptively learn the Kriging proxy model until the U learning function convergence condition is met. After the introduction of SISPDF, the failure probability P f The calculation expression is:
[0092]
[0093] In order to ensure the convergence of the calculation results, 50 groups of failure probability estimates are estimated based on the established Kriging model according to formula (3) through 50 different samplings. Calculate the corresponding mean and variance, and then according to Get the coefficient of variation of the estimated value To determine whether the final Kriging proxy model is acceptable.
[0094] After the surrogate model constructed by AK combined with SISPDF efficiently calculates the failure probability with a small number of samples, the quasi-Monte Carlo (QMC) theory is introduced into the calculation of the reliability sensitivity index of the input variable under the distribution parameter uncertainty, and the entire calculation process is further simplified to a single-layer method. This is similar to the single-layer method in the article "Efficient Reliability Sensitivity Analysis Method under Parameter Uncertainty", and the unconditional variance V(I F )as well as and Then the main sensitivity index S is calculated i and total sensitivity index
[0095] Below, reference Figure 1 As shown in , the above steps are described in more detail.
[0096] Step 1: Generate Θ X The following is a sample matrix of the distribution parameters, which is denoted by Θ.
[0097] Expand Θ to get
[0098] Step 2: Extraction of sample pool. Extract N of the input variable X X Group Sample As the sample pool S for constructing the Kriging model IS .
[0099] Step 3: Construct the initial Kriging proxy model. Mpp Substitute the input-output samples into the original function g(X) and solve for the corresponding function value This forms the initial training set T IS According to T IS Construct the initial Kriging model g k (X).
[0100] Step 4: Select the sample update point x' u . Using the current g k (X), calculate the sample pool S IS The U learning function value of each sample point in is then Select the updated sample point x' u .
[0101] Step 5: Determine whether the adaptive learning process has converged. u )≥2, stop the adaptive learning process and execute step 6. Otherwise, calculate x' u The actual model output value g(x' u ), and (x' u ,g(x' u )) Add to the current training set, that is, T IS ={T IS ∪(x' u ,g(x' u ))}, return to step 3 to continue updating the Kriging model.
[0102] Step 6: Solve the failure probability and its coefficient of variation. Using the current Kriging proxy model g k (X) Prediction sample pool S IS The failure domain indicator function I of all samples in F (x), and the estimated failure probability is obtained according to formula (3): Get different failure probability estimates through 50 different samplings and calculate the corresponding means With variance And calculate the corresponding coefficient of variation when We think P f The estimate is acceptable, and the final Kriging proxy model g is obtained. k (X); if Then expand the sample pool S IS , and go to step three.
[0103] Step 7: Generate samples of X. Generate independent N×n dimensional sample matrices A and B of the input variable X. Where:
[0104]
[0105]
[0106] Assign the elements of matrix A to C (i) , except for the i-th column element, we get
[0107]
[0108] Step 8: Substitute matrices A, B, and C (i) Substitute the samples in into the Kriging surrogate model g k (X), the corresponding output value is obtained: Y kA =g k (A), Y kB =g k (B) Then determine whether the output value falls within the failure domain F = {x:g(x)≤0} and find the corresponding indicator function: I FA =I F (A), I FB =I F (B)
[0109] Step 9: Calculate the main sensitivity index S of the input variable X according to equations (1) and (2): i and total sensitivity index At this time, calculate the main sensitivity index S i and total sensitivity index The specific expression is:
[0110] and
[0111]
[0112] Denominator V(I F ) is calculated as:
[0113]
[0114] in,
[0115]
[0116] Example 2 AKSL-TIS method for calculating input variable reliability sensitivity index
[0117] On the basis of the AKSL-IS method, in order to further save calculation samples, the present invention studies a single-layer parameter uncertainty reliability sensitivity analysis (AKSL-TIS) method based on an adaptive Kriging model combined with a surrogate truncated importance sampling probability density function (STISPDF).
[0118] The main idea of truncated importance sampling (TIS) is to define a standard normal space, take the origin as the center of the circle, and use the obtained reliability index as the radius of the circle to draw a hypersphere. It is stipulated that the points falling inside the space are safe and can be deleted without calculation. In this way, only the function values of the points outside the sphere need to be calculated, which ensures the accuracy of this method while further reducing the amount of calculation. Let g(X) be an n-dimensional independent standard normal variable X = {X1, X2, …, X n} T In the independent standard normal space, the reliability index β is the shortest distance from the origin to the limit state surface g(x) = 0, that is: Obviously, in the n-dimensional independent standard normal space, the equation Define the n-dimensional Euclidean space R n The n-dimensional hypersphere with the origin as the center and β as the radius is usually called the β-sphere. The β-sphere divides the space R n Divide into ||x|| 2 <β 2 and ||x|| 2 ≥β 2 Two parts. Since β is the shortest distance from the origin to the limit state surface g(x) = 0, ||x|| 2 <β 2 The determined area must be within the safety domain S = {x:g(x)>0}. By combining the proxy sampling probability density function with the truncated importance sampling method, a proxy truncated importance sampling probability density function (STISPDF) can be constructed.
[0119] AKSL-TIS can perform reliability analysis by calculating the performance function values corresponding to important sampling samples falling outside the β-sphere. 2 ≥β 2 The failure domain F = {x|g(x)≤0} is included, so we introduce the region ||x|| 2 ≥β 2 The indicator function I β (x) The calculation formula for the failure probability obtained by the truncated importance sampling method is:
[0120]
[0121] The failure probability estimate is obtained by using the Kriging model constructed by the above formula Calculate the corresponding mean and variance, and then get the coefficient of variation of the estimated value according to formula (4): To determine whether the final Kriging proxy model is acceptable.
[0122] Similarly, the surrogate model constructed by AK combined with STISPDF can efficiently calculate the failure probability with a small number of samples, and then the main sensitivity index S can be calculated. i and total sensitivity index
[0123] Below, reference Figure 1 As shown in , the above steps are described in more detail.
[0124] Step 1: Generate Θ X First, extract the sample matrix of the distribution parameters, the sample matrix is recorded as Θ, that is,
[0125] Expand Θ to get
[0126] Step 2: Extraction of sample pool. Extract N of the input variable X X Group Sample As the sample pool S for constructing the Kriging model TIS .
[0127] Step 3: Construct the initial Kriging proxy model. Mpp Substitute the input-output samples into the original function g(X) and solve for the corresponding function value This forms the initial training set T TIS According to T TIS Construct the initial Kriging model g k (X).
[0128] Step 4: Select the sample update point x' u . Using the current g k (X), calculate the sample pool S TIS The U learning function value of each sample point in is then calculated according to Select the updated sample point x' u .
[0129] Step 5: Determine whether the adaptive learning process has converged. u )≥2, stop the adaptive learning process and execute step 6. Otherwise, calculate x' u The actual model output value g(x' u ), and (x' u ,g(x' u )) Add to the current training set, that is, T TIS ={T TIS ∪(x' u ,g(x' u ))}, return to step 3 to continue updating the Kriging model.
[0130] Step 6: Solve the failure probability and its coefficient of variation. Using the current Kriging proxy model g k (X) Prediction sample pool S TIS The failure domain indicator function I of all samples in F (x), and the estimated failure probability is obtained according to formula (4): Get different failure probability estimates through 50 different samplings and calculate the corresponding means With variance And calculate the corresponding coefficient of variation when We think P f The estimate is acceptable, and the final Kriging proxy model g is obtained. k (X); if Then expand the sample pool S TIS , and go to step three.
[0131] Step 7: Generate samples of X. Generate independent N×n dimensional sample matrices A and B of the input variable X. Where:
[0132]
[0133]
[0134] Assign the elements of matrix A to C (i) , except for the i-th column element, we get:
[0135]
[0136] Step 8: Use the β-sphere to transform the matrices A, B, and C (i) The samples falling into the safe region are cut off, and the corresponding indicator function is obtained: I βA =I β (A), I βB =I β (B)
[0137] Step 9: Reshape the sample points whose β-ball indicator function is not 0 into a new matrix, denoted as A', B', C' (i) , and then through the newly fitted Kriging surrogate model g k (X) Calculate the value of the new matrix output function, that is, the corresponding input-output value: Y′ kA =g k (A′), Y′ kB =g k (B′),
[0138] Step 10: After obtaining the output value of the new matrix, estimate whether these sample points are failed and obtain the corresponding failure domain indicator function: I′ FA =I F (A′),I′ FB =I F (B′),
[0139] Step 11, discard the sample points whose beta ball indicator function is 0. There is no need to calculate the performance function value again. Instead, the sample points can be directly replaced with 0 in the matrix of the failure domain indicator function, thereby forming a new matrix.
[0140] Step 12: Calculate the main sensitivity index S of the input variable X according to equations (1) and (2): i and total sensitivity index At this time, calculate the main sensitivity index S i and total sensitivity index The specific expression is: and
[0141] Denominator V(I F ) is calculated as:
[0142]
[0143] in,
[0144]
[0145] The working process of the present invention is as follows:
[0146] The analysis results of the headless rivet riveting process example using the present invention are as follows:
[0147] In aerospace engineering, headless rivets are usually used for riveting. There are many factors that affect the quality of riveting during the process, among which extrusion stress is the most important factor. Therefore, it is necessary to control the influence of extrusion stress during the riveting process. In order to establish the relationship between extrusion stress and rivet size, several ideal conditions are assumed:
[0148] (1) The rivet hole does not expand during the riveting process;
[0149] (2) The change in rivet volume can be ignored;
[0150] (3) After riveting, the rivet head is cylindrical;
[0151] (4) The materials are isotropic.
[0152] Before riveting, the initial volume V0 of the rivet can be expressed as:
[0153]
[0154] Where d and h are the diameter and height of the rivet in state A respectively.
[0155] After stage I, the volume of the rivet in state B can be expressed as:
[0156]
[0157] Where D0 and h1 are the diameter and height of the rivet in state B respectively.
[0158] After stage II, assuming that the upper and lower parts of the rivet in state C have the same size, the volume in state C can be expressed as:
[0159]
[0160] Where t is the overall thickness of the thin-walled part, D1 and H are the diameter and height of the rivet head in state C, respectively.
[0161] According to the hardening strength theory, the maximum extrusion stress in the y direction can be expressed as:
[0162]
[0163] Where K is the intensity factor, n SHE is the hardening factor of the rivet material, ε y is the true strain of the rivet head during the riveting process. yIt consists of two parts: the strain of the nail rod in the upsetting stage (stage I) and the strain in the upsetting forming stage (stage II) So the true strain ε y It can be expressed as:
[0164]
[0165] In the formula,
[0166] Assuming that the volume of the rivet remains unchanged during the riveting process, the maximum extrusion stress of the rivet can be obtained by rearranging the above formula:
[0167]
[0168] The rivet material selected in the present invention is 2017-T4, and its hardening index n SHE =0.15, upsetting head height H = 2.2mm. If the maximum extrusion stress is greater than the extrusion strength, the rivet will fail, so the function is established as follows:
[0169] g=σ sq -σ max (11)
[0170] In this example, we give the compression strength of the rivet as σ sq =580MPa, in the whole riveting process, the size of the rivet and the characteristics of the material can be considered as normal random variables, and the distribution parameters are shown in Table 1. It is assumed that the mean of each input variable is also a normal variable, and the parameters are shown in Table 2.
[0171] Table 1 Distribution parameters of rivet structure input variables
[0172]
[0173] Table 2 Distribution parameters of input variable means
[0174]
[0175] Table 3 Sensitivity analysis results of headless rivets based on reliability
[0176]
[0177]
[0178] Table 4 Reliability analysis results of various methods
[0179]
[0180] See also Figure 4The embodiment of the present invention proposes an inexact probability input reliability sensitivity analysis system based on an adaptive Kriging model, comprising:
[0181] An initial Kriging model building module extracts an initial training sample set from the distribution parameters of the target structure based on a sampling probability density function, and builds an initial Kriging model based on the initial training sample set;
[0182] An adaptive learning module performs adaptive learning on the initial Kriging model based on the U learning function; when the adaptive learning process converges, the failure probability and its coefficient of variation are solved; otherwise, the initial Kriging model is rebuilt based on the initial training sample set until the adaptive learning process converges;
[0183] An adaptive Kriging model construction module, when the coefficient of variation meets the requirements, an adaptive Kriging model is obtained; otherwise, the initial training sample set is expanded, and the initial Kriging model is reconstructed based on the expanded initial training sample set until its coefficient of variation meets the requirements;
[0184] An indicator function calculation module generates a sample matrix of input variables based on a sampling probability density function, substitutes samples in the sample matrix into the adaptive Kriging model, and obtains corresponding output values and their indicator functions;
[0185] The sensitivity analysis module calculates the main sensitivity index and the total sensitivity index of the target structure based on the indicator function, and performs reliability sensitivity analysis.
[0186] The present invention discloses an inexact probability input reliability sensitivity analysis system based on an adaptive Kriging model, which combines the adaptive Kriging model with a sampling probability density function, replaces the implicit and complex limit functional state function in actual engineering with an explicit and simpler model, and shows significant superiority in the global reliability sensitivity analysis of inexact probability input. Furthermore, the sampling probability density function includes the importance sampling probability density function (SISPDF) and the truncated importance sampling probability density function (STISPDF). The adaptive Kriging model is combined with the importance sampling probability density function (SISPDF) and the truncated importance sampling probability density function (STISPDF). The efficient methods based on the adaptive Kriging model are proposed respectively: the single-layer parameter uncertainty reliability sensitivity analysis based on the adaptive Kriging model combined with SISPDF (AKSL-IS) method and the single-layer parameter uncertainty reliability sensitivity analysis based on the adaptive Kriging model combined with STISPDF (AKSL-TIS) method. The example results show that the AKSL-IS and AKSL-TIS methods update the important SS-PDF, update the sampling center through iterative design points, so that the sample points are as far as possible in the failure domain, reduce the sample pool size, and improve the calculation efficiency. The AKSL-TIS method introduces the β-sphere indicator function, and by screening the sample points that fall into the β-sphere, there is no need to calculate their model output values, so that the calculation efficiency is further improved. In addition, for extremely complex limit state functions or implicit functional functions, it takes a long time for MCS to call the model once. By embedding the adaptive Kriging proxy model to convert it into a simple explicit function, reliability analysis can be performed more quickly and efficiently.
[0187] like Figure 5 As shown, the third object of the present invention is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the reliability sensitivity analysis method of inexact probability input based on the adaptive Kriging model when executing the computer program.
[0188] The reliability sensitivity analysis method of inexact probability input based on the adaptive Kriging model comprises the following steps:
[0189] Extracting an initial training sample set from the distribution parameters of the target structure based on a sampling probability density function, and constructing an initial Kriging model based on the initial training sample set;
[0190] Adaptively learning the initial Kriging model based on the U learning function; when the adaptive learning process converges, solving the failure probability and its coefficient of variation, otherwise, reconstructing the initial Kriging model based on the initial training sample set until its adaptive learning process converges;
[0191] When the coefficient of variation meets the requirement, an adaptive Kriging model is obtained; otherwise, the initial training sample set is expanded, and the initial Kriging model is reconstructed based on the expanded initial training sample set until its coefficient of variation meets the requirement;
[0192] Generate a sample matrix of input variables based on a sampling probability density function, substitute samples in the sample matrix into the adaptive Kriging model, and obtain corresponding output values and their indicator functions;
[0193] The main sensitivity index and the total sensitivity index of the target structure are calculated based on the indicator function, and reliability sensitivity analysis is performed.
[0194] The fourth object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the steps of the reliability sensitivity analysis method of inexact probability input based on the adaptive Kriging model.
[0195] The reliability sensitivity analysis method of inexact probability input based on the adaptive Kriging model comprises the following steps:
[0196] Extracting an initial training sample set from the distribution parameters of the target structure based on a sampling probability density function, and constructing an initial Kriging model based on the initial training sample set;
[0197] Adaptively learning the initial Kriging model based on the U learning function; when the adaptive learning process converges, solving the failure probability and its coefficient of variation, otherwise, reconstructing the initial Kriging model based on the initial training sample set until its adaptive learning process converges;
[0198] When the coefficient of variation meets the requirement, an adaptive Kriging model is obtained; otherwise, the initial training sample set is expanded, and the initial Kriging model is reconstructed based on the expanded initial training sample set until its coefficient of variation meets the requirement;
[0199] Generate a sample matrix of input variables based on a sampling probability density function, substitute samples in the sample matrix into the adaptive Kriging model, and obtain corresponding output values and their indicator functions;
[0200] The main sensitivity index and the total sensitivity index of the target structure are calculated based on the indicator function, and reliability sensitivity analysis is performed.
[0201] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may 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.
[0202] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0203] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0204] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0205] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A reliability sensitivity analysis method for inexact probability input based on adaptive Kriging model, characterized in that: The following steps are involved: Extracting an initial training sample set from the distribution parameters of the target structure based on a sampling probability density function, and constructing an initial Kriging model based on the initial training sample set; Performing adaptive learning on the initial Kriging model based on the U learning function; When the adaptive learning process converges, the failure probability and its coefficient of variation are solved, otherwise, the initial Kriging model is rebuilt based on the initial training sample set until the adaptive learning process converges; When the coefficient of variation meets the requirement, an adaptive Kriging model is obtained; otherwise, the initial training sample set is expanded, and the initial Kriging model is reconstructed based on the expanded initial training sample set until its coefficient of variation meets the requirement; Generate a sample matrix of input variables based on a sampling probability density function, substitute samples in the sample matrix into the adaptive Kriging model, and obtain corresponding output values and their indicator functions; The main sensitivity index and the total sensitivity index of the target structure are calculated based on the indicator function, and reliability sensitivity analysis is performed.
2. The reliability sensitivity analysis method for inexact probability input based on adaptive Kriging model according to claim 1 is characterized in that: The step of extracting an initial training sample set from the distribution parameters of the target structure based on the sampling probability density function specifically includes: Extract the sample matrix of the distribution parameters of the target structure; Based on the sampling probability density function, several groups of samples of the input variables are extracted as the sample pool; A number of input-output samples in the process of solving the design point based on the sample matrix and the sample pool are brought into the original function, and a number of groups of function values are obtained as initial training sample sets.
3. The reliability sensitivity analysis method for inexact probability input based on adaptive Kriging model according to claim 2 is characterized in that: The adaptive learning of the initial Kriging model based on the U learning function and judging whether the adaptive learning process converges specifically include: Calculate the U learning function value of each sample point in the sample pool based on the initial Kriging model; Select and update the sample points according to the U learning function value of each sample point to obtain the U learning sample points; When the U learning function value of the U learning sample point is ≥2, the adaptive learning process converges.
4. The reliability sensitivity analysis method for inexact probability input based on adaptive Kriging model according to claim 3 is characterized in that: When the adaptive learning process does not converge, reconstructing the initial Kriging model based on the initial training sample set until the adaptive learning process converges includes: The actual model output value of the U learning sample point is calculated, and the coordinate point composed of the U learning sample point and its actual model output value is added to the initial training sample set, and the initial Kriging model is reconstructed until its adaptive learning process converges.
5. The reliability sensitivity analysis method for inexact probability input based on adaptive Kriging model according to claim 4 is characterized in that: The solution of failure probability and its coefficient of variation specifically includes: Predicting the failure domain indicator function of all samples in the sample pool based on the initial Kriging model, thereby obtaining a failure probability estimate as the failure probability; Through several samplings, several failure probability estimates are obtained and their means and variances are calculated, and then the corresponding coefficient of variation is calculated.
6. The reliability sensitivity analysis method for inexact probability input based on adaptive Kriging model according to claim 5 is characterized in that: When the coefficient of variation meets the requirement, an adaptive Kriging model is obtained; Expanding the initial training sample set and reconstructing the initial Kriging model based on the expanded initial training sample set until its coefficient of variation meets the requirements specifically includes: When the coefficient of variation is less than 5%, an adaptive Kriging model is obtained; When the coefficient of variation is ≥5%, the initial training sample set is expanded, and the initial Kriging model is reconstructed based on the expanded initial training sample set until the coefficient of variation meets the requirement.
7. The reliability sensitivity analysis method for inexact probability input based on adaptive Kriging model according to claim 1 is characterized in that: The sampling probability density function includes an importance sampling probability density function and a truncated importance sampling probability density function.
8. The reliability sensitivity analysis method for inexact probability input based on adaptive Kriging model according to claim 7 is characterized in that: The generating of a sample matrix of input variables based on a sampling probability density function, substituting samples in the sample matrix into the adaptive Kriging model, and obtaining corresponding output values and their indicator functions specifically include: When the sampling probability density function is an important sampling probability density function: Generate independent N×n dimensional sample matrices A and B of the input variables based on the importance sampling probability density function: Assign the elements of matrix A to C (i) , except for the i-th column element, we get: The matrices A, B and C (i) Substitute the samples in into the adaptive Kriging model and obtain the corresponding output value: Y kA =g k (A), Y kB =g k (B) Then determine whether the output value falls within the failure domain F = {x:g(x)≤0} and find the corresponding indicator function: I FA =I F (A),I FB =I F (B), 9. The reliability sensitivity analysis method for inexact probability input based on adaptive Kriging model according to claim 8 is characterized in that: The generating of a sample matrix of input variables based on a sampling probability density function, substituting samples in the sample matrix into the adaptive Kriging model, and obtaining corresponding output values and their indicator functions specifically include: When the sampling probability density function is a truncated importance sampling probability density function, the following steps are also included: The matrices A, B and C are transformed by β-sphere (i) The samples falling into the safe region are cut off, and the corresponding indicator function is obtained: I βA =I β (A), I βB =I β (B) The sample points whose β-ball indicator function is not 0 are re-formed into a new matrix, denoted as A', B', C' (i) , and then calculate the value of the new matrix output function through the newly fitted adaptive Kriging model, that is, the corresponding input-output value: Y k ' A =g k (A'), Y k ' B =g k (B'), After obtaining the output values of the new matrix, estimate whether these sample points are failed and obtain the corresponding failure domain indicator function: I' FA =I F (A'),I' FB =I F (B'), By discarding the sample points whose beta-sphere indicator function is 0, there is no need to calculate their performance function values again. Instead, they can be directly replaced with 0 at the original positions in the matrix of the failure domain indicator function, thereby forming a new matrix.
10. An inexact probability input reliability sensitivity analysis system based on an adaptive Kriging model, characterized in that: include: An initial Kriging model building module extracts an initial training sample set from the distribution parameters of the target structure based on a sampling probability density function, and builds an initial Kriging model based on the initial training sample set; An adaptive learning module, which performs adaptive learning on the initial Kriging model based on a U learning function; When the adaptive learning process converges, the failure probability and its coefficient of variation are solved, otherwise, the initial Kriging model is rebuilt based on the initial training sample set until the adaptive learning process converges; An adaptive Kriging model construction module, when the coefficient of variation meets the requirements, an adaptive Kriging model is obtained; otherwise, the initial training sample set is expanded, and the initial Kriging model is reconstructed based on the expanded initial training sample set until its coefficient of variation meets the requirements; An indicator function calculation module generates a sample matrix of input variables based on a sampling probability density function, substitutes samples in the sample matrix into the adaptive Kriging model, and obtains corresponding output values and their indicator functions; The sensitivity analysis module calculates the main sensitivity index and the total sensitivity index of the target structure based on the indicator function, and performs reliability sensitivity analysis.
Citation Information
Cited By
Reliability analysis method based on adaptive agent model and importance sampling
CN120145735A
Engineering structure reliability analysis method based on extended Kriging agent model
CN120781663A
