A global sensitivity analysis method for the reliability of a lattice sandwich structure
By using a Gaussian process regression model and the Metropolis-Hastings algorithm to correct intersection points in the reliability analysis of lattice sandwich structures, the problem of insufficient analysis accuracy in traditional methods is solved, and more accurate global sensitivity analysis of reliability is achieved.
Patent Information
- Application Number
- CN202611600734.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-16
- Publication Date
- 2026-11-17
AI Technical Summary
In the global sensitivity analysis of the reliability of lattice sandwich structures, the traditional three-point second-order interpolation method is difficult to approximate the highly nonlinear limit state function, which causes the intersection point to deviate from the true failure boundary, reducing the accuracy of the analysis, especially in the tail region of the failure domain where the error accumulation is serious.
By transforming the influencing parameters to the standard normal space, a Gaussian process regression model is used to predict the limit state function value. The Metropolis-Hastings algorithm is then used to sample the posterior distribution, generating a Markov chain to correct the intersection points. This constructs a corrected set of reliability indicators, improving the accuracy of the analysis.
It accurately covers the real failure boundaries of the lattice structure under all operating conditions, avoids result distortion, and improves the accuracy of global reliability sensitivity analysis.
Smart Images

Figure CN123116506A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of structural reliability assessment and uncertainty quantification technology, and in particular to a global sensitivity analysis method for the reliability of a lattice sandwich structure. Background Technology
[0002] Lattice sandwich structures, with their excellent properties such as lightweight, high specific strength, high specific stiffness, and designability, are widely used in components such as cabin bulkheads. Additive manufacturing technology is the mainstream process for fabricating lattice sandwich structures. However, in the layer-by-layer additive manufacturing process, uncertainties such as uneven powder distribution, laser power fluctuations, differences in scanning strategies, and temperature field evolution inevitably lead to defects in lattice sandwich structures, such as geometric dimensional deviations and material property dispersion. The coupling effect of these uncertainties directly affects the load-bearing capacity and service safety of the structure; therefore, it is necessary to conduct reliability assessments during the structural design phase. Among these, global reliability sensitivity analysis can quantitatively reveal the contribution of each random parameter to the probability of structural failure, providing important decision-making basis for the identification of key parameters, structural optimization design, and process parameter control.
[0003] Existing technologies typically perform global sensitivity analysis based on adaptive line sampling. This method determines important directions by solving for the most likely failure point; several sampling lines are generated along the important directions, and three initial coefficients are selected on each sampling line. The intersection of each sampling line with the failure boundary (the failure boundary is the boundary surface between the reliable domain and the failure domain of the lattice sandwich structure) is calculated using the three-point second-order interpolation method to obtain the reliability index and then estimate the failure probability.
[0004] However, when the adaptive line sampling method solves for the intersection of each sampling line with the failure boundary, the quadratic polynomial constructed by the traditional three-point second-order interpolation method often fails to approximate the highly nonlinear true limit state function, causing the solved intersection points to deviate significantly from the true failure boundary. Especially when the intersection point is located at the tail of the failure domain, the existing technology only uses the mean of the reliability indices of other sampling lines for rough filling, which reduces the accuracy of global reliability sensitivity analysis. Summary of the Invention
[0005] The purpose of this invention is to provide a global sensitivity analysis method for the reliability of lattice sandwich structures, which improves the accuracy of global sensitivity analysis for the reliability of lattice sandwich structures.
[0006] To address the aforementioned technical problems, embodiments of the present invention provide a method for global reliability sensitivity analysis of a lattice sandwich structure, comprising the following steps: Obtain the influence parameters of the lattice sandwich structure model, transform the influence parameters to the standard normal space, and solve for the most likely failure point of the lattice sandwich structure in the standard normal space. Sample points are drawn from a standard normal distribution, and sampling lines are constructed for each sample point along the direction of the most likely failure point. The intersections of each sampling line with the failure boundary are initially calculated to obtain a preliminary reliability index set. Multiple high-error intersections with interpolation residuals greater than a preset threshold are identified from the preliminary reliability index set as a prior set to be corrected. Each intersection to be corrected in the prior set is considered as a current state variable, and a likelihood function is constructed conditioned on the actual failure surface. The posterior distribution is determined based on the likelihood function and the predetermined prior distribution. The posterior distribution is sampled using the Metropolis-Hastings algorithm to generate a Markov chain. The mean of the samples on the Markov chain is used as the correction value for the intersection to be corrected. A corrected reliability index set is constructed based on the correction values of all intersections to be corrected. The global reliability sensitivity index of the lattice sandwich structure is calculated based on the revised reliability index set.
[0007] Optionally, the most likely failure point of the lattice sandwich structure in the standard normal space is determined by the following steps: Obtain a coarse estimate of the most likely failure point in the first layer of a lattice sandwich structure in standard normal space; Centered on the coarse estimate of the most likely failure point in the first layer, multiple candidate samples following a normal distribution are extracted; Using the Gaussian process regression model trained in the first layer, the limit state function value at each candidate sample in the first layer is predicted, and the predicted limit state function values are sorted in ascending order to obtain the ranking results; the Gaussian process regression model corresponding to each layer is trained based on the approximate distance function with probability weighting and the expected improvement function; The predicted values of the limit state function of the first preset number of samples are selected from the sorting results and used as the failure threshold of the second layer. By performing layered iterations until the failure threshold meets the termination condition, the most likely failure point of the corresponding layer when the termination condition is met is determined as the most likely failure point of the lattice sandwich structure in the standard normal space.
[0008] Optionally, obtaining a coarse estimate of the most likely failure point of the first layer of the lattice sandwich structure in standard normal space includes the following steps: Randomly select sample points in the standard normal space and calculate the response value of the sample points using the true limit state function; The training set is formed by combining sample points and their corresponding response values. The initial Gaussian regression model is trained using the training set, and the parameters of the initial Gaussian regression model are updated using the expectation improvement criterion. After convergence, a coarse estimate of the most likely failure point of the first layer is obtained.
[0009] Optionally, the first lThe method for determining the failure threshold of a layer includes the following steps: With the first The most likely failure point of the layer Establish a hyperrectangular feasible region centered on the target; Within the hyperrectangular feasible region, extract multiple samples that follow the mean value. Candidate samples with a normal distribution and a predefined variance; Using the first The pre-trained Gaussian process regression model predicts the limit state function value at each candidate sample and sorts the predicted limit state function values in ascending order. Select the predicted value of the limit state function of the nth sample as the nth sample. l Layer failure threshold.
[0010] Optionally, the first The approximate distance function for layered probability weighting is: ; in, For the first Approximate distance function of the failure surface in the middle of the layer. To predict the standard deviation, For the first l Layer failure threshold, Let be a vector in standard normal space, and || || be the Euclidean norm of the vector. To act as a proxy for the Gaussian process regression model, For indicator functions, It is a minimal positive constant. For soft boundary coefficients, For the first l Failure threshold for layer -1; No. The expected improvement function of the layer is: ; ; in, For the first The expected improvement function value of the failure surface in the middle of the layer. For the first The current optimal solution of the layer, The representation vector is located at the th l The confidence level of the layer failure surface and the condition that meets the soft boundary constraints; The probability that the current optimal solution is determined by the Gaussian process model to be located near the failure boundary of this layer and satisfying the soft boundary constraint. To predict the mean, For error tolerance, It is the standard normal cumulative distribution function.
[0011] Optionally, the process of constructing the limit state function includes the following steps: Obtain the influence parameters of the lattice sandwich structure model; Construct a limit state function to describe the failure of a lattice sandwich structure; Transform the influencing parameters to the standard normal space to obtain the limit state function in the standard normal space.
[0012] Optionally, the predetermined prior distribution is a uniform distribution.
[0013] Optionally, the global reliability sensitivity index of the lattice sandwich structure is calculated based on the modified reliability index set, including the following steps: Substitute the modified reliability index set into the standard normal cumulative distribution function to calculate the arithmetic mean of all modified reliability indices; The arithmetic mean is used to determine the failure probability of the lattice sandwich structure. The marginal density functions of each influencing parameter are estimated using the kernel probability density estimation method. Based on the failure probability of the lattice sandwich structure and the edge density function of each influencing parameter, the global reliability sensitivity index of each influencing parameter in the lattice sandwich structure is calculated.
[0014] Optionally, based on the failure probability of the lattice sandwich structure and the edge density function of each influencing parameter, the global reliability sensitivity index of each influencing parameter in the lattice sandwich structure is calculated. The calculation process is expressed as follows: ; in, For the first i A global sensitivity index for each influencing parameter. This represents the failure probability of the lattice sandwich structure. Indicates the input variable The mathematical expectation, Under parameter failure conditions The conditional marginal probability density function, for The marginal probability density, |.| denotes taking the absolute value.
[0015] Optionally, the influencing parameters include rod diameter, elastic modulus, load, and Poisson's ratio.
[0016] A global sensitivity analysis device for the reliability of a lattice sandwich structure, specifically comprising: The solution module is used to obtain the influence parameters of the lattice sandwich structure model, transform the influence parameters to the standard normal space, and solve for the most likely failure point of the lattice sandwich structure in the standard normal space. The correction module is used to extract sample points from the standard normal distribution and construct sampling lines for each sample point along the direction of the most likely failure point; it initially solves for the intersection points of each sampling line with the failure boundary to obtain a preliminary reliability index set; it identifies multiple high-error intersection points with interpolation residuals greater than a preset threshold from the preliminary reliability index set as a prior set to be corrected, and treats each intersection point to be corrected in the prior set as the current state variable, constructing a likelihood function conditioned on the actual failure surface; it determines the posterior distribution based on the likelihood function and the predetermined prior distribution, samples the posterior distribution using the Metropolis-Hastings algorithm to generate a Markov chain, and uses the mean of the samples on the Markov chain as the correction value for the intersection point to be corrected; and it constructs a corrected reliability index set based on the correction values of all intersection points to be corrected. The determination module is used to calculate the global reliability sensitivity index of the lattice sandwich structure based on the modified reliability index set.
[0017] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described global sensitivity analysis method for the reliability of the lattice sandwich structure.
[0018] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the above-described global sensitivity analysis method for the reliability of the lattice sandwich structure.
[0019] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: In the global sensitivity analysis method for the reliability of lattice sandwich structures provided in this specification, the present invention uses multiple high-error intersection points with interpolation residuals greater than a preset threshold identified in the preliminary reliability index as intersection points to be corrected. A likelihood function conditioned on the real failure surface is constructed, and a posterior target distribution is constructed by combining the prior distribution. A Markov chain is generated by sampling using the Metropolis-Hastings algorithm, and the mean of the samples on the chain is used as the intersection point correction value. The generated Markov chain sample sequence will eventually converge to the pre-constructed target posterior distribution. In this way, all the sampled samples strictly follow the probability distribution conditioned on the real failure surface, and there will be no sampling bias. This allows the corrected reliability index set to accurately cover the real failure boundary of the lattice structure under all working conditions, avoiding the result distortion caused by the simplification of failure modes in conventional methods, thereby improving the accuracy of the global sensitivity analysis of the reliability of lattice sandwich structures. Attached Figure Description
[0020] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.
[0021] Figure 1 This is a flowchart illustrating the global sensitivity analysis method for the reliability of a lattice sandwich structure described in this specification. Figure 2 This manual provides a simplified diagram and a structural loading diagram of the lattice sandwich structure. Figure 2 (a) in the diagram is a schematic diagram of the lattice sandwich structure. Figure 2 (b) is a simplified diagram of the lattice sandwich structure; Figure 3 This is a schematic diagram of mesh generation in finite element analysis provided in this manual; Figure 4 This is a stress distribution cloud diagram provided in this manual for the given parameters in the finite element analysis. Figure 5 This is a bar chart of the global sensitivity index provided in this manual; Figure 6 This is a schematic diagram of a global sensitivity analysis device for the reliability of a lattice sandwich structure provided in this specification; Figure 7 This is a schematic diagram of a computer device provided in this specification for performing a global reliability sensitivity analysis of a lattice sandwich structure. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.
[0023] Devices such as desktop computers, servers, and laptops are capable of executing the solutions described in this manual. For ease of explanation, the following description will focus on servers as the primary execution method.
[0024] Adaptive line sampling is a commonly used method for handling high-dimensional, nonlinear problems with low failure probabilities. Its basic process is as follows: First, important directions are determined by solving for the most likely failure point (i.e., the design point). Second, several sampling lines are generated along these important directions, with three initial coefficients selected for each line. A three-point second-order interpolation method is used to calculate the intersection of each sampling line with the failure boundary, obtaining a reliability index, and then estimating the failure probability. Since directly calling the true limit state function is costly, a Gaussian process regression model is introduced as a surrogate model, and an adaptive learning function is constructed for model updates. During the iteration process, the learning function actively selects the sample points that contribute most to the prediction uncertainty to update the surrogate model. The trained Gaussian process model efficiently predicts the limit state function values and intersections on each sampling line, gradually approximating the true failure boundary, significantly reducing the number of calls to the original high-fidelity model while maintaining computational accuracy. Based on this, failure samples are sampled from the standard normal distribution and selected according to the reliability index for each sampling line. The marginal probability density of the calculated parameters under failure conditions is estimated using kernel density, thus obtaining the global sensitivity index.
[0025] In this process, the accuracy of the design points and the precision of the intersection point calculation determine the reliability of failure probability estimation and sensitivity analysis. While adaptive line sampling methods have proven capable of handling such complex problems, their performance suffers from two key limitations: first, their performance depends on the accurate solution of the design points; second, due to the nonlinearity and multiple extrema characteristics of the model, as well as the mismatch in the interpolation interval, the quadratic polynomial constructed by the traditional three-point second-order interpolation method often fails to approximate the true function shape when calculating the intersection points of each sampling line with the failure boundary. This results in the solved intersection points deviating significantly from the true failure boundary, rendering the solution ineffective. To address this, traditional methods pre-fill the intersection points using the mean reliability index of other sampling lines. However, this substitution strategy is prone to significant cumulative errors in the tail region, reducing the accuracy of failure probability estimation and sensitivity analysis. Therefore, compensating for the inaccurate solution of design points and intersection points in traditional adaptive line sampling methods is crucial for global reliability sensitivity analysis.
[0026] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.
[0027] Figure 1 This is a flowchart illustrating the global sensitivity analysis method for the reliability of a lattice sandwich structure described in this specification, which specifically includes the following steps: S101. Obtain the influence parameters of the lattice sandwich structure model, transform the influence parameters to the standard normal space, and solve for the most likely failure point of the lattice sandwich structure in the standard normal space.
[0028] In an exemplary embodiment, the process of finding the most probable failure point of a lattice sandwich structure in standard normal space includes the following steps: obtaining a coarse estimate of the first-layer most probable failure point of the lattice sandwich structure in standard normal space; extracting multiple candidate samples that follow a normal distribution, centered on the coarse estimate of the first-layer most probable failure point; using the Gaussian process regression model trained in the first layer to predict the limit state function value at each candidate sample in the first layer, and arranging the predicted limit state function values in ascending order to obtain the ranking result; the trained Gaussian process regression model corresponding to each layer is trained based on the approximate distance function with probability weighting and the expected improvement function; selecting the predicted limit state function values of the first preset number of samples from the ranking result as the failure threshold of the second layer; and performing layered iterations until the failure threshold meets the termination condition, and determining the most probable failure point of the corresponding layer when the termination condition is met as the most probable failure point of the lattice sandwich structure in standard normal space.
[0029] In an exemplary embodiment, obtaining a coarse estimate of the most likely failure point in the first layer of a lattice sandwich structure in a standard normal space includes the following steps: randomly sampling sample points in the standard normal space and calculating the response value of the sample points using the true limit state function; combining the sample points and their corresponding response values into a training set; training an initial Gaussian regression model using the training set and updating the parameters of the initial Gaussian regression model using the expected improvement criterion; and obtaining a coarse estimate of the most likely failure point in the first layer after convergence.
[0030] In one exemplary embodiment, the first l The method for determining the failure threshold of a layer includes the following steps: With the first The most likely failure point of the layer Establish a hyperrectangular feasible region centered at [center]; within the hyperrectangular feasible region, extract multiple samples that follow the mean [value]. Candidate samples with a normal distribution and a predefined variance; using the first... The pre-trained Gaussian process regression model predicts the limit state function value at each candidate sample, and sorts the predicted limit state function values in ascending order; the predicted limit state function value of the i-th sample is selected as the j-th sample. l Layer failure threshold.
[0031] Specifically, the preset number of iterations and preset variance are set according to specific engineering practices.
[0032] In an exemplary embodiment, the process of constructing the limit state function includes the following steps: obtaining the influence parameters of the lattice sandwich structure model; constructing the limit state function to describe the failure of the lattice sandwich structure; and transforming the influence parameters to the standard normal space to obtain the limit state function in the standard normal space.
[0033] No. The approximate distance function for the layer is given by formula (1): (1); in, For the first Approximate distance function of the failure surface in the middle of the layer. To predict the standard deviation, For the first l Layer failure threshold, Let be a vector in standard normal space, and || || be the Euclidean norm of the vector. To act as a proxy for the Gaussian process regression model, For indicator functions, It is a minimal positive constant. For soft boundary coefficients, For the first l The failure threshold for layer -1.
[0034] No. The expected improvement function of the layer is given by formula (2): (2); in, , For the first The expected improvement function value of the failure surface in the middle of the layer. For the first The current optimal solution of the layer, The representation vector is located at the th l The confidence level for the layer failure surface and the condition satisfying soft boundary constraints. The probability that the current optimal solution is determined by the Gaussian process model to be located near the failure boundary of this layer and satisfying the soft boundary constraint. To predict the mean, For error tolerance, It is the standard normal cumulative distribution function.
[0035] Specifically, firstly, the influence parameters and their probability distributions of the lattice sandwich structure model are obtained; given the failure conditions, a limit state function describing the structural failure is constructed. The input variables in the original space are transformed using an equal probability transformation method. Transforming to standard normal space yields the limit state function in standard normal space. ,in z Following a standard normal distribution, the failure domain is defined as follows. .
[0036] To address the problem that traditional methods for finding the most likely failure point easily get trapped in local optima under conditions of high nonlinearity and low failure probability, a Bayesian subset optimization algorithm is introduced to solve for the design point. Most likely failure point. It is the point with the shortest distance from the origin to the boundary of the failure domain in the standard normal space. Define the minimum positive constant. Define the distance function ,in, For indicator functions, when hour, Otherwise, it is zero. The maximum value of the distance function is the most likely failure point. Solving for the most likely failure point is transformed into an optimization problem of finding the maximum value of the distance function. Bayesian optimization is used to quickly obtain a coarse estimate of the most likely failure point. A small number of initial samples are randomly selected from the standard normal space, and the response value is calculated using the true limit state function. The small number of sample points and their corresponding response values constitute the initial training set, and an initial Gaussian process model is trained. A surrogate model is used to proxy the distance function. Then, based on the expected improvement criterion, the potential expectation of candidate sample points in improving the distance function value compared to the current optimal solution is quantified. New sample points are iteratively selected to update the model until convergence, obtaining the estimate of the first layer of most likely failure points. .
[0037] Introducing a series of intermediate failure thresholds Its corresponding intermediate failure surface is ,in, b l Gradually decreasing until approaching 0, the failure threshold is progressively approximated to the most likely failure point. The failure threshold for each layer is calculated based on the design points of the layer above. First, it is necessary to use the design points of the previous layer as a basis... Layer design points Establish a system with a center and a side length of... ( Take the feasible region of the hyperrectangle (2~4). Let be the length of any side of the hyperrectangular feasible region. Within this hyperrectangular feasible region, extract... K The mean is 1. A sample that follows a normal distribution with a variance of 1. Using the... The trained Gaussian process model predicts the limit state function value at each candidate sample. The predicted values are sorted in ascending order, and the first value is selected. The predicted value of the nth sample is used as the nth l Layer failure threshold , The selection coefficient for a small proportion of samples is usually... p 0 is set between 0.01 and 0.05. (If the failure threshold...) Then terminate the iteration and simultaneously let The most likely failure point of this layer As the most likely point of failure.
[0038] In each layer, a locally accurate Gaussian process model needs to be trained to identify the most likely failure point for that layer. Due to the uncertainty in the predictions of the Gaussian process model, its prediction standard deviation is large when a certain region has not been fully sampled, and the model may underestimate or overestimate the true limit state function value. If a traditional hard boundary condition is used... The strategy is to discard points whose predicted values are temporarily higher than a threshold but may actually be within the failure domain. This can cause the algorithm to prematurely abandon potentially valuable regions, especially when the design point is far from the origin and the failure surface is highly nonlinear, easily leading the search into local optima. The invention introduces a soft boundary mechanism: in defining the... l When using the distance function of the layer, it is required that the candidate sample points satisfy the condition that the predicted limit state value is close to the current threshold. And it also satisfies the condition that the predicted value does not exceed the expansion range of the upper-level boundary (i.e. ,in, α This is the soft boundary coefficient, with a value ranging from 0.5 to 1. (The standard deviation of the model prediction). After using the Gaussian process model to surrogate the failure surface of the current layer, the approximate distance function shown in formula (1) is constructed. Based on this distance function, the improved expected improvement function is constructed as shown in formula (2). This design makes the optimization objective no longer the deterministic reciprocal of the distance, but the probability-weighted expected distance, taking into account the misjudgment caused by the uncertainty of prediction. The active learning training process of each layer of Gaussian process model is as follows: with the current design point A candidate sample pool is generated within a hyperrectangular region with sides of length 2–4 at the center. A Gaussian regression model is trained based on this sample pool until the maximum value of the desired improvement function is less than a preset threshold Δ. EI2 (e.g. 10) -6 The optimal solution predicted by the current model is taken as the most likely failure point of this layer. .
[0039] Iterate continuously until the failure threshold is reached. Then let The most likely failure point of this layer As the most likely point of failure The most likely failure point obtained from the final optimization. The direction is an important direction for line sampling. An adaptive linear sampling method is used to initially solve for the intersection points, which are drawn from the standard normal distribution. sample points Along important directions Parallel lines are constructed using a Gaussian process regression model trained in Bayesian subset optimization as the initial surrogate model, which is continuously updated in subsequent calculations through an active learning strategy. Specifically, for the current Gaussian process model, the learning function for all candidate sample points is calculated based on the predicted mean and variance. The point with the smallest learning function value is selected, and its response is calculated by calling the true limit state function. This point is then added to the training set, and the Gaussian process model is retrained. This process is repeated until the learning function values for all candidate points are greater than a preset stopping threshold. Using a fully trained model, a three-point quadratic interpolation method is used to initially solve for the intersection points corresponding to each sample point, obtaining preliminary reliability indices for all parallel sampling lines. For intersection points that cannot be calculated using the three-point second-order interpolation method, the mean of the reliability index of the remaining sample points is used instead.
[0040] In one exemplary embodiment, the influencing parameters include rod diameter, elastic modulus, load, and Poisson's ratio.
[0041] In one exemplary embodiment, the lattice sandwich structure analyzed in this embodiment is made of GH4169 material and fabricated using selective laser melting (SLM) additive manufacturing. The overall dimensions of the structure are 22.5mm × 25.5mm × 29mm, with each unit cell employing a body-centered cubic (BCC) structure, measuring 4.5mm × 4.5mm × 4.5mm, and an average rod diameter of 0.9354mm. A fixed constraint is applied to the lower surface of the structure, while a downward compressive load is applied to the top surface. Figure 2 This manual provides a simplified diagram and a structural loading diagram of the lattice sandwich structure. Figure 2 (a) in the diagram is a schematic diagram of the lattice sandwich structure. Figure 2 (b) is a simplified diagram of the lattice sandwich structure.
[0042] Consider the four types of uncertainties affecting the structural mechanical properties: rod diameter r Elastic modulus E Load P and Poisson's ratio ν The distribution types and statistical characteristics of each influencing parameter are shown in Table 1. The failure criterion for the structure is that the average displacement of the top surface of the sandwich structure does not exceed the failure displacement of 0.8821 mm (obtained from quasi-static compression tests), and the limit state function is constructed accordingly. ,in, The average displacement of the top surface is obtained by finite element simulation calculation under given parameters.
[0043] Table 1 Characterization of Uncertainty Parameters
[0044] After constructing the limit state function model, the most likely failure point is first solved using the Bayesian subset optimization method. The various influencing parameters are then... , The diameter of the rod. For elastic modulus, For load, Given the Poisson ratio, transform to the standard normal space, denoted as . . Figure 3 This is a schematic diagram of mesh generation in finite element analysis provided in this manual. Figure 4 This is a stress distribution contour map provided in this manual for the finite element analysis with given parameters. The initial training sample size is 12, and the initial sample set is generated in standard normal space using the Latin hypercube sampling method. The finite element simulation program is then called to calculate the limit state function value corresponding to each sample. G ( z The specific mesh generation and stress distribution cloud map under specific parameters for finite element simulation are shown in the following figures. Figure 3 and Figure 4 As shown, an initial Gaussian process regression model is trained. The kernel function is chosen as a squared exponential kernel, and the hyperparameters are determined through maximum likelihood estimation.
[0045] To solve for the most likely failure point using Bayesian subset optimization, the algorithm parameters are first set, and the soft boundary coefficient is taken. α It takes the value 1, a very small positive constant. =0.05, the expected stopping threshold Δ of the improved function. EI2 Set to 10 -5 intermediate probability p 0=0.01, the side length of the hyperrectangle is set to 4, used to determine the number of candidate samples for the failure threshold. K =20000.
[0046] The coarse estimate of the most likely failure point in the first layer was obtained by calculating using the Bayesian optimization method. The following is the first... l The failure threshold of a layer is determined as follows: based on the current most likely failure point. Extract from a hyperrectangular neighborhood with a side length of 4 centered on the target. A sample (subject to the following) (A normal distribution centered at x with a standard deviation of 1) using an existing Gaussian process model (i.e., the 1st... l The model trained with layer -1 predicts the limit state value for each sample, sorts all the limit state values in ascending order, and takes the nth value. The predicted value of each sample is used as the failure threshold. .
[0047] In the lIn the active learning process of the layer, a soft boundary mechanism is introduced to account for the uncertainty of the Gaussian process prediction. Candidate sample points are defined. z The predicted value is close to the current failure threshold. b l The probability function that does not exceed the boundary of the previous layer is given by formula (3): (3); in, Candidate sample points z The predicted value is close to the current failure threshold. b l And the probability value does not exceed the boundary value of the previous layer. For error tolerance, The standard normal cumulative distribution function is... For coefficients, For candidate sample points z The standard deviation of the Gaussian process prediction, For candidate sample points z The mean predicted by the Gaussian process.
[0048] The improved expected improvement function is constructed as shown in Equation (2). The particle swarm optimization algorithm is used to maximize the expected improvement function over the entire feasible region. The optimal candidate point is obtained. This point is added to the training set, and the Gaussian process model is retrained until the desired improvement function converges, thus obtaining the final design point for this layer. Iterate continuously until the failure threshold is reached. In this embodiment, after three iterations, the most important direction is finally obtained. .
[0049] S102. Sample points are drawn from the standard normal distribution, and sampling lines are constructed for each sample point along the direction of the most likely failure point. The intersection points of each sampling line and the failure boundary are initially solved to obtain a preliminary reliability index set. Multiple high-error intersection points with interpolation residuals greater than a preset threshold are identified from the preliminary reliability index set as a prior set to be corrected. Each intersection point to be corrected in the prior set is regarded as the current state variable, and a likelihood function is constructed with the actual failure surface as the condition. The posterior distribution is determined according to the likelihood function and the predetermined prior distribution. The posterior distribution is sampled using the Metropolis-Hastings algorithm to generate a Markov chain. The mean of the samples on the Markov chain is used as the correction value of the intersection point to be corrected. The corrected reliability index set is constructed based on the correction values of all intersection points to be corrected.
[0050] In one exemplary embodiment, the predetermined prior distribution is a uniform distribution.
[0051] After obtaining the preliminary reliability indices, the predicted standard deviation of each intersection point in the preliminary reliability indices is further calculated. Points with higher predicted standard deviations are selected to form a priori set to be corrected. These intersection points are usually located in the tail region of the limit state surface, and their distribution deviates from the actual failure boundary. A Markov chain method is used for accurate correction. For each intersection point in the priori set to be corrected, a Markov chain is run independently, and the intersection point is corrected to the correct posterior distribution position through MetropolisHastings sampling.
[0052] The specific process is as follows: The Gaussian process regression model, which has already been trained using adaptive linear sampling, is used as a proxy, and its predicted mean is... The prediction standard deviation is Define the log-likelihood function. , Let be the likelihood function value. The larger the likelihood function value, the more likely the point is to lie on the true failure surface. The target distribution of a Markov chain is defined as the product of the likelihood function and the prior distribution, where the prior distribution is uniform (no information prior). Therefore, the probability density of the posterior distribution is: , For a predetermined prior distribution, For the posterior distribution, Intersection coefficients to be corrected c The likelihood function at that location.
[0053] An improved adaptive linear sampling method is adopted, combined with the corrected precise intersection points, to generate a failure sample set. Specifically, this includes: constructing multiple sampling lines using the adaptive linear sampling method; generating candidate samples on each sampling line according to a standard normal distribution; filtering the candidate samples using a corrected reliability index set, retaining samples with reliability indices greater than the reliability index as failure samples on the sampling line; and integrating all failure samples on the sampling lines into a failure sample set.
[0054] Initial values for each intersection point to be corrected The MetropolisHastings sampling step is executed independently. First, initialization is performed, setting the current state variable to the intersection value obtained from the initial solution. Calculate the log-likelihood value of the current state. A symmetric Gaussian proposal distribution is used, with a step size of 0.5. The proposed step size can be adaptively adjusted during sampling to maintain the acceptance rate at an optimal level (around 0.44). Given the sampling length for each chain... T Generate candidate samples from the proposal distribution Calculate the likelihood value of the candidate samples. Since the prior distribution is uniform, the prior density ratio is 1, and the acceptance rate simplifies to... Random numbers are drawn from a uniform distribution U(0,1). If the random number is less than the acceptance rate... If a candidate sample is found to be positive, it is accepted; otherwise, it is rejected. To determine whether the Markov chain has reached a stationary distribution, multiple independent chains (e.g., 4-6) are run simultaneously. The Gelman-Rubin diagnostic metric is used for convergence judgment: the sample variance within each chain is calculated. Inter-chain variance Thus, diagnostic volume is obtained. When diagnostic volume When the value approaches 1 (usually less than 1.05), the algorithm is considered converged. Otherwise, the number of chains is increased, and parameter adjustments continue. Finally, for each intersection, the mean of the samples after the chains have stabilized is taken as the corrected intersection value, thus achieving Bayesian correction of the initial intersection.
[0055] In the direction of the most likely failure point Then, an adaptive linear sampling method is used to initially solve for the intersection points corresponding to each sample point, thus obtaining the initial reliability index. Samples are drawn from the standard normal distribution. sample points , M This represents the number of sample points. For each sample point, it is decomposed into parallel and perpendicular components along the significance direction: ,in, To and Parallel coordinates, The component perpendicular to the important direction. Fixed vertical component. By changing c The value will give you a straight line parallel to the important direction.
[0056] For the j For each sample, select three initial values for the coefficients along its corresponding straight line. Predict the limit state function values at these three points using the current Gaussian process model. Then, the response curve on the straight line is fitted using the three-point quadratic interpolation method, and the curve and the failure surface are solved. The intersection of these points yields the reliability index of the sample. .
[0057] The active learning process in this process is as follows: In the initial stage, the Gaussian process model trained in step two is used to calculate... The learning function value at each candidate sample point ,in, The representative mean is variance is The probability density of the Gaussian distribution, This represents the estimate of the limit state function by the Gaussian process model. The variance of the Gaussian process model is represented by... The mean of the Gaussian process model is represented by... The error tolerance is used to control the width of the integration interval, and its value should be selected close to the function value at the intersection point. The learning function is increased by 10-100 times. The points with the smallest learning function among all sample points are added to the training set of the Gaussian process. This process is iterated until the learning function values at all sample points exceed the stopping threshold. Updates are stopped. The pre-trained Gaussian model is used to calculate the intersections of all sample point sampling lines with the failure boundary. For sample points where the intersection calculation has issues, the reliability index of the remaining sample point sampling lines is used. The mean value is used instead. In this embodiment, the active learning process calls the real simulation model a total of 65 times. After completing the above process, all results can be obtained. M Preliminary reliability index set of 2000 samples .
[0058] After obtaining the initial reliability index, the interpolation residuals at each intersection point are calculated, and points with high residuals are selected to form the prior set to be corrected. For each intersection point with a large interpolation residual, a Markov chain is run independently, and the residuals are corrected using Metropolis Hastings sampling. The Gaussian process regression model trained in step three is used to surrogate the true limit state function, and its predicted mean is... The prediction standard deviation is For each intersection point to be corrected Let its corresponding sample point be . The vertical component along the important direction is Let the corrected intersection point be... The corresponding coordinates of the failure sample are The predicted limit state function value of the failure sample point follows a normal distribution. Define the log-likelihood function. Let the prior distribution be a uniform distribution. Then the posterior target distribution is proportional to the product of the likelihood and the prior distribution: .
[0059] For each initial value of the intersection point to be corrected, the MetropolisHastings sampling step is executed independently. First, initialization is performed, setting the current state variable to the intersection point value obtained from the initial solution. Calculate the log-likelihood value of the current state. A symmetric Gaussian proposal distribution is used, with a proposal step size of 0.5. To maintain sampling efficiency, the step size is dynamically adjusted every 100 steps based on the actual acceptance rate, with a target acceptance rate of 0.44. The sampling length for each chain is given. Generate candidate samples from the proposal distribution Calculate the coordinates of the failure points of the candidate samples. And predicted by the Gaussian process model and This allows for the calculation of the likelihood values of candidate samples. Calculate the reception probability Random numbers are drawn from a uniform distribution U(0,1). r If random number r Less than the probability of receiving α If a candidate sample is found to be positive, it is accepted; otherwise, it is rejected. To determine whether the Markov chain has reached a stationary distribution, the MH algorithm is run multiple times, and the variance ratio between multiple chains is calculated. This variance ratio is used to determine whether the chain has reached the stationary state of the target distribution. The Gelman-Rubin diagnostic metric is used for convergence judgment: the sample variance within each chain is calculated. W MH Inter-chain variance B MH Then the total variance estimate is Thus, diagnostic volume is obtained. .in This represents the number of samples for each chain. In this embodiment, all high-error intersections... If all values are less than 1.05, it is determined that all chains have converged to the same posterior distribution.
[0060] Replace the original values of all high-error points with their respective corrected values, while leaving the remaining intersection points unchanged, to form a complete set of corrected reliability indices. This metric is used for subsequent sensitivity calculations.
[0061] For each sampling line, a large number of candidate samples are generated according to the standard normal distribution. The reliability index corrected using this sampling line Filter and retain all that meet the criteria. The candidate samples are used as the failed samples on this sampling line. (Summary) The failure samples from the sampling lines constitute the failure sample set. The kernel probability density estimation method is used to estimate the first... i Marginal probability density of each variable Estimate the failure condition of the first i Marginal probability density of each variable .
[0062] S103. Calculate the global reliability sensitivity index of the lattice sandwich structure based on the revised reliability index set.
[0063] In an exemplary embodiment, the global reliability sensitivity index of the lattice sandwich structure is calculated based on the modified reliability index set, including the following steps: substituting the modified reliability index set into the standard normal cumulative distribution function to calculate the arithmetic mean of all modified reliability indices; determining the arithmetic mean as the failure probability of the lattice sandwich structure; estimating the marginal density function of each influencing parameter using the kernel probability density estimation method; and calculating the global reliability sensitivity index of each influencing parameter in the lattice sandwich structure based on the failure probability of the lattice sandwich structure and the marginal density function of each influencing parameter.
[0064] Define the concept of global sensitivity index: Let the influence parameter vector be... Its joint probability density function is , No. i The marginal probability density function of each influencing parameter is: The limit state function is The failure domain is The failure probability is .
[0065] In an exemplary embodiment, the global reliability sensitivity index of each influencing parameter in the lattice sandwich structure is calculated based on the failure probability of the lattice sandwich structure and the edge density function of each influencing parameter. The calculation process is expressed as formula (4): (4); in, For the first i A global sensitivity index for each influencing parameter. This represents the failure probability of the lattice sandwich structure. Indicates the input variable The mathematical expectation, Under parameter failure conditions The conditional marginal probability density function, for The marginal probability density, |.| denotes taking the absolute value.
[0066] Calculate the failure probability of a lattice sandwich structure Coefficient of variation of failure probability and sensitivity index The calculated failure probability and sensitivity index results of this embodiment are shown in Table 2. In this embodiment, the calculated failure probability is... Coefficient of variation of failure probability The coefficient of variation is much less than 0.05, indicating that the estimation results of the failure probability are convergent and have a high degree of confidence.
[0067] Updating the reliability of lattice sandwich structures requires calling numerous simulation models, which traditional reliability analysis methods cannot efficiently solve. Therefore, the IM-AGPR-LS and AK-MCS algorithms are compared. To ensure robustness and accuracy, and to keep the coefficient of variation of the failure probability below 5%, the AK-MCS algorithm uses 400,000 sample points to calculate the failure probability, while the IM-AGPR-LS algorithm uses 2,000 sample points. (Table...) This represents the number of times the limit state function is called. The number in parentheses represents the total number of samples, and the number in parentheses represents the standard deviation of the results of 10 independent and repeated experiments.
[0068] Table 2 Reliability and Sensitivity Analysis Results
[0069] The failure probability and global sensitivity analysis results of the lattice sandwich structure were analyzed, and the failure probability and sensitivity ranking of the algorithms in the table are consistent. Active Learning Kriging Monte Carlo Simulation (AK-MCS) and Improved Adaptive Global Polynomial Regression Local Sampling Method were used. The computational costs of the two algorithms (Sampling and IM-AGPR-LS) are 605 and 65, respectively. The IM-AGPR-LS method can obtain the global sensitivity at a very low computational cost when facing complex low-probability, nonlinear problems, while the AK-MCS method is more costly, proving the high accuracy and efficiency of the algorithm proposed in this invention. The sensitivity analysis index results are slightly different, but the sensitivity ranking is consistent. Among them, the rod diameter has the greatest impact on the reliability of the lattice sandwich structure, followed by load, elastic modulus, and Poisson's ratio. This result shows that the sensitivity ranking of parameters changes significantly after considering model correction, and rod diameter and load become important factors. The computational cost of the AK-MCS method is 605, while the computational efficiency of the IM-AGPR-LS method is several times higher, showing extremely high computational efficiency. It can obtain the same sensitivity ranking as AK-MCS with only a few samples, which is especially suitable for the rapid solution of low-probability, high-dimensional nonlinear problems. Finally, under 10 independent repeated experiments, the standard deviation of the failure probability and the sensitivity index is small, indicating that the calculation results are stable.
[0070] Figure 5 This is a bar chart of the global sensitivity index provided in this manual, such as... Figure 5As shown, for the significantly influential variables of rod diameter, elastic modulus, and load, the exponential results of the three algorithms are nearly consistent. The relative errors of IM-AGPR-LS and AK-MCS are both controlled within ±6%, verifying the credibility of the proposed surrogate model in sensitivity quantification. Among them, the sensitivity indices of rod diameter and load are significantly higher than those of other variables, indicating that rod diameter and load are the main influencing parameters of displacement response, followed by elastic modulus, while Poisson's ratio has a negligible impact.
[0071] Based on the above method, the present invention provides a global reliability sensitivity analysis system for lattice sandwich structures based on improved adaptive line sampling, including: a model building module, a design point solving module, an intersection point solving and correction module, and a sensitivity calculation module.
[0072] The model building module is used to obtain the influence parameters and their probability distribution of the structural model, construct the limit state function, and transform the input variables to the standard normal space. The design point solution module is used to solve for the design points in the standard normal space using the Bayesian subset optimization algorithm, and uses the direction of the design point as an important direction for line sampling. Specifically, the Bayesian subset optimization constructs a series of intermediate failure surfaces and introduces a soft boundary mechanism, actively training the Gaussian process regression model using the expected improvement function to approximate the true design point layer by layer. The intersection point solution and correction module is used to extract sample points from the standard normal distribution, use an adaptive line sampling method to initially solve for the intersection points corresponding to each sample point, obtain preliminary reliability indices, and identify... Intersections with solution biases are identified and form a prior set to be corrected. Then, the Metropolis-Hastings algorithm from the Markov chain Monte Carlo method is used, with a standard normal distribution as the target distribution, to perform conditional sampling and posterior correction on the intersections to be corrected, resulting in the corrected reliability index. The sensitivity calculation module is used to calculate the failure probability based on the corrected reliability index. Candidate samples are generated along each sampling line according to the standard normal distribution, and failure samples are selected using the corresponding reliability index. The failure sample set is then compiled, and the marginal probability density function under the failure conditions of each input variable is calculated using kernel probability density estimation. Finally, the global sensitivity index is calculated, and the analysis results are output.
[0073] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the technical scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention. Furthermore, the division of the above method steps is only for the sake of clarity of description. In actual implementation, multiple steps can be combined into one step, or one step can be split into multiple sub-steps, as long as they contain the same logical relationship, they all fall within the scope of protection of the present invention. Non-substantial modifications, parameter adjustments, or the introduction of irrelevant designs to the algorithm or process, if they do not change its core idea and key processes, should also be considered to fall within the scope of protection of the present invention.
[0074] The above describes a global sensitivity analysis method for the reliability of a lattice sandwich structure provided by one or more embodiments of this specification. Based on the same idea, this specification also provides a corresponding global sensitivity analysis device for the reliability of a lattice sandwich structure, such as... Figure 6 As shown.
[0075] Figure 6 A schematic diagram of a global sensitivity analysis device for the reliability of a lattice sandwich structure provided in this specification includes: The solver module 601 is used to obtain the influence parameters of the lattice sandwich structure model, transform the influence parameters to the standard normal space, and solve for the most likely failure point of the lattice sandwich structure in the standard normal space.
[0076] The correction module 602 is used to extract sample points from the standard normal distribution and construct sampling lines for each sample point along the direction of the most likely failure point; initially solve the intersection points of each sampling line with the failure boundary to obtain a preliminary reliability index set; identify multiple high-error intersection points with interpolation residuals greater than a preset threshold from the preliminary reliability index set as a prior set to be corrected; treat each intersection point to be corrected in the prior set as the current state variable and construct a likelihood function conditioned on the actual failure surface; determine the posterior distribution based on the likelihood function and the predetermined prior distribution; sample the posterior distribution using the Metropolis-Hastings algorithm to generate a Markov chain; use the mean of the samples on the Markov chain as the correction value of the intersection point to be corrected; and construct the corrected reliability index set based on the correction values of all intersection points to be corrected.
[0077] The determination module 603 is used to calculate the global reliability sensitivity index of the lattice sandwich structure based on the modified reliability index set.
[0078] Specific limitations regarding the global sensitivity analysis device for the reliability of lattice sandwich structures can be found in the limitations of the global sensitivity analysis method for lattice sandwich structures described above, and will not be repeated here. Each module in the aforementioned global sensitivity analysis device for the reliability of lattice sandwich structures can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.
[0079] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 A global sensitivity analysis method for the reliability of lattice sandwich structures is provided.
[0080] This instruction manual also provides Figure 7 The schematic diagram of the computer device shown is as follows: Figure 7 At the hardware level, the computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 A global sensitivity analysis method for the reliability of lattice sandwich structures is provided.
[0081] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.
[0082] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
Claims
1. A global sensitivity analysis method for the reliability of a lattice sandwich structure, characterized in that, The method includes the following steps: Obtain the influence parameters of the lattice sandwich structure model, transform the influence parameters to the standard normal space, and solve for the most likely failure point of the lattice sandwich structure in the standard normal space. Sample points are drawn from a standard normal distribution, and sampling lines are constructed for each sample point along the direction of the most likely failure point. The intersections of each sampling line with the failure boundary are initially calculated to obtain a preliminary reliability index set. Multiple high-error intersections with interpolation residuals greater than a preset threshold are identified from the preliminary reliability index set as a prior set to be corrected. Each intersection in the prior set to be corrected is considered as a current state variable, and a likelihood function is constructed conditioned on the actual failure surface. The posterior distribution is determined based on the likelihood function and the predetermined prior distribution. The posterior distribution is sampled using the Metropolis-Hastings algorithm to generate a Markov chain. The mean of the samples on the Markov chain is used as the correction value for the intersection to be corrected. A corrected reliability index set is constructed based on the correction values of all intersections to be corrected. The global reliability sensitivity index of the lattice sandwich structure is calculated based on the revised reliability index set.
2. The global sensitivity analysis method for the reliability of the lattice sandwich structure as described in claim 1, characterized in that, The process of finding the most likely failure point of a lattice sandwich structure in standard normal space includes the following steps: Obtain a coarse estimate of the most likely failure point in the first layer of a lattice sandwich structure in standard normal space; Centered on the coarse estimate of the most likely failure point in the first layer, multiple candidate samples following a normal distribution are extracted; Using the Gaussian process regression model trained in the first layer, the limit state function value at each candidate sample in the first layer is predicted, and the predicted limit state function values are sorted in ascending order to obtain the ranking results; the Gaussian process regression model corresponding to each layer is trained based on the approximate distance function with probability weighting and the expected improvement function; The predicted limit state function values of the first preset number of samples from the sorting results are selected as the failure threshold of the second layer. By performing layered iterations until the failure threshold meets the termination condition, the most likely failure point of the corresponding layer when the termination condition is met is determined as the most likely failure point of the lattice sandwich structure in the standard normal space.
3. The global sensitivity analysis method for the reliability of the lattice sandwich structure as described in claim 2, characterized in that, The process of obtaining a coarse estimate of the most likely failure point of the first layer of a lattice sandwich structure in standard normal space includes the following steps: Randomly select sample points in the standard normal space and calculate the response value of the sample points using the true limit state function; The training set is formed by combining sample points and their corresponding response values. The initial Gaussian regression model is trained using the training set, and the parameters of the initial Gaussian regression model are updated using the expected improvement criterion. After convergence, a coarse estimate of the most likely failure point of the first layer is obtained.
4. The global sensitivity analysis method for the reliability of the lattice sandwich structure as described in claim 2, characterized in that, No. l The method for determining the failure threshold of a layer includes the following steps: With the first The most likely failure point of the layer Establish a hyperrectangular feasible region centered on the target area; Within the hyperrectangular feasible region, multiple samples are extracted that follow the mean value. Candidate samples with a normal distribution and a predefined variance; Using the first The trained Gaussian process regression model predicts the limit state function value at each candidate sample and sorts the predicted limit state function values in ascending order. Select the predicted value of the limit state function of the nth sample as the nth sample. l The failure threshold of the layer.
5. The global sensitivity analysis method for the reliability of the lattice sandwich structure as described in claim 2, characterized in that, No. The approximate distance function for layered probability weighting is: ; in, For the first Approximate distance function of the failure surface in the middle of the layer. To predict the standard deviation, For the first l Layer failure threshold, Let be a vector in standard normal space, and || || be the Euclidean norm of the vector. To act as a proxy for the Gaussian process regression model, For indicator functions, It is a minimal positive constant. For soft boundary coefficients, For the first l Failure threshold for layer -1; No. The expected improvement function of the layer is: ; ; in, For the first The expected improvement function value of the failure surface in the middle of the layer. For the first The current optimal solution of the layer, The representation vector is located at the th l The confidence level of the layer near the failure surface and satisfying soft boundary constraints; The probability that the current optimal solution is determined by the Gaussian process model to be located near the failure boundary of this layer and satisfying the soft boundary constraint. To predict the mean, For error tolerance, It is the standard normal cumulative distribution function.
6. The global sensitivity analysis method for the reliability of the lattice sandwich structure as described in claim 2, characterized in that, The process of constructing the limit state function includes the following steps: Obtain the influence parameters of the lattice sandwich structure model; Construct a limit state function to describe the failure of a lattice sandwich structure; Transform the influencing parameters to the standard normal space to obtain the limit state function in the standard normal space.
7. The global sensitivity analysis method for the reliability of the lattice sandwich structure as described in claim 1, characterized in that, The predetermined prior distribution is a uniform distribution.
8. The global sensitivity analysis method for the reliability of the lattice sandwich structure as described in claim 1, characterized in that, The calculation of the global reliability sensitivity index of the lattice sandwich structure based on the modified reliability index set includes the following steps: Substitute the modified reliability index set into the standard normal cumulative distribution function to calculate the arithmetic mean of all modified reliability indices; The arithmetic mean is determined as the failure probability of the lattice sandwich structure; The marginal density functions of each influencing parameter are estimated using the kernel probability density estimation method. Based on the failure probability of the lattice sandwich structure and the edge density function of each influencing parameter, the global reliability sensitivity index of each influencing parameter in the lattice sandwich structure is calculated.
9. The global sensitivity analysis method for the reliability of the lattice sandwich structure as described in claim 8, characterized in that, The global reliability sensitivity index of each influencing parameter in the lattice sandwich structure is calculated based on the failure probability of the lattice sandwich structure and the edge density function of each influencing parameter. The calculation process is as follows: ; in, For the first i A global sensitivity index for each influencing parameter. This represents the failure probability of the lattice sandwich structure. Indicates the input variable The mathematical expectation, Under parameter failure conditions The conditional marginal probability density function, for The marginal probability density, |.| denotes taking the absolute value.
10. The global sensitivity analysis method for the reliability of the lattice sandwich structure as described in claim 1, characterized in that, The influencing parameters include rod diameter, elastic modulus, load, and Poisson's ratio.
