Bearing structure accelerated storage degradation assessment method based on adaptive Kriging proxy modeling

Abstract

The invention provides a load-bearing structure accelerated storage degradation evaluation method based on self-adaptive Kriging proxy modeling, which is characterized in that degradation modeling is carried out on material-level mechanical properties of a load-bearing structure based on an accelerated degradation evaluation theory, and material-level degradation information is further transmitted to a structure level by combining structural mechanical simulation response analysis and a proxy modeling theory, so that the load-bearing structure accelerated storage degradation evaluation method based on self-adaptive Kriging proxy modeling is realized. The method is used for carrying out structure-level degradation modeling and evaluation of a load-bearing structure, and solves the problems that a traditional method of the load-bearing structure is low in evaluation precision and difficult to carry out evaluation. The method comprises the following steps of: 1, estimating parameters of a material-level accelerated degradation model; 2, carrying out structure-level simulation test design and initial modeling; 3, carrying out self-adaptive iterative updating and convergence judgment on the structure-level model; and 4, evaluating the degradation of the structure-level bearing capacity. The method is suitable for accelerated storage reliability evaluation of the load-bearing structure, does not need to carry out a large number of structure-level accelerated storage tests, and has the characteristics of high efficiency and low cost.

Description

Technical Field

[0001] The present invention relates to a method for accelerating the storage reliability assessment of a load-bearing structure based on an adaptive Kriging surrogate model. It is a method for assessing the storage reliability of a load-bearing structure based on the Kriging surrogate model and the accelerated degradation assessment theory. For the accelerated storage test data of the mechanical properties of the load-bearing structure, based on the accelerated degradation assessment theory, a degradation model of the mechanical properties of the load-bearing structure is established and used as the input, and an adaptive response analysis of the structural mechanics simulation is carried out to establish a surrogate model of the structural bearing performance, and finally the storage reliability of the load-bearing capacity of the load-bearing structure is evaluated. It is applicable to fields such as the accelerated storage degradation assessment and response modeling analysis of load-bearing structures. Background Art

[0002] Generally, there are some structures with certain load-bearing capacities inside the missile and rocket products, such as the cabin structure. After mass production, the missile and rocket products need to be stored in the warehouse for a period of time, which is called the storage period of the product. During the storage period, due to the action of the long-term storage environment, the different performances of the load-bearing structures of the missile and rocket products may degenerate. In order to identify the changes in the mechanical properties of the load-bearing structure during the long storage period, it is generally necessary to carry out accelerated degradation tests to evaluate and model the performance of the structure during the storage period. Considering the extremely high cost of the overall accelerated test of the structure, accelerated storage tests are often carried out on its material-level performance.

[0003] After obtaining the storage degradation models of different performances of the load-bearing structure materials, how to obtain the degradation model of the overall structural bearing performance based on the material-level performance degradation model is a recognized difficult problem in the field of accelerated degradation assessment. Conventional system integration methods mainly transfer component-level degradation information to the system level based on simple reliability block diagram relationships. However, for the hierarchical relationship between materials and structures with complex physical laws, the reliability block diagram is difficult to represent the internal correlation relationship. In order to characterize the influence of materials on the structural response, it is generally necessary to carry out modeling through simulation tests. Considering the computational time cost of the simulation model, a surrogate modeling method can be used to improve the analysis efficiency. Adaptive Kriging is an effective surrogate modeling technique and has a good balance in terms of accuracy and efficiency.

[0004] Based on this, the present invention proposes a method for accelerating the storage reliability assessment of a load-bearing structure based on the accelerated degradation assessment theory and the Kriging model, comprehensively considering the material-level accelerated test data and the structural-level simulation data, and giving the parameters of the material-level accelerated degradation model, the structural-level accelerated degradation model and the storage reliability assessment results. Summary of the Invention

[0005] The object of the present invention is to provide an accelerated storage reliability assessment method for load-bearing structures in view of the problem that traditional system integration methods cannot be used for the storage degradation assessment of load-bearing structures. It is a structural-level accelerated storage degradation assessment method that includes material-level accelerated degradation modeling, structural-level simulation test design and initial modeling, structural-level model adaptive iterative update and convergence discrimination, and structural-level load-bearing capacity degradation assessment. Based on the accelerated degradation assessment theory, the mechanical properties of the load-bearing structure are degraded and modeled as input, and the adaptive response analysis of the structural mechanics simulation is carried out to establish a surrogate model of the structural load-bearing performance, and finally the storage reliability of the load-bearing capacity of the load-bearing structure is evaluated.

[0006] To achieve the above object, an accelerated storage degradation assessment method for load-bearing structures based on adaptive Kriging surrogate modeling of the present invention needs to establish the following basic settings:

[0007] Setting 1: The storage degradation law of the structural material follows one of the linear degradation, exponential degradation, and power function degradation models, and the expressions of each model are as follows:

[0008] ① Linear degradation model:

[0009] y(t) = α + β·t + σ·ε (1)

[0010] ② Exponential degradation model:

[0011] y(t) = μ·exp(λ·t) + σ·ε (2)

[0012] ③ Power function degradation model:

[0013] y(t) = η·t θ + σ·ε (3)

[0014] Where t represents the storage time, α and β represent the parameters of the linear degradation model, μ and λ represent the parameters of the exponential degradation model, η and θ represent the parameters of the power function degradation model, ε represents a standard normal distribution variable with a mean of 0 and a variance of 1, and σ is the standard deviation.

[0015] Setting 2: β, λ, and η in the above three degradation models are all acceleration-related parameters, and we uniformly denote them as θ. Considering that the acceleration law of the structural material follows one of the Arrhenius model and the inverse power law model, the corresponding model expressions are:

[0016] Arrhenius model:

[0017] lnθ = a + b / S (4)

[0018] Inverse power law model:

[0019] lnθ = a + b ln S (5)

[0020] Among them, θ represents the acceleration-related parameter in the degradation model, a and b represent the acceleration model parameters, and S represents the acceleration stress.

[0021] Setting 3: The simulation model of the structure is known, that is, when inputting the m-dimensional performance parameter y, the response value of the structure's bearing capacity can be obtained. Define the input-output relationship as:

[0022] z = G(y) (6)

[0023] Among them, z represents the response value of the structure's bearing capacity, and G(y) represents the response model.

[0024] The Kriging model can be used to substitute the above input-output relationship, that is:

[0025] G(y) = β0 + z G (y) (7)

[0026] Among them, β0 is a constant representing the global trend, and z G (y) is a stationary Gaussian process with zero mean and covariance σ 2 r(y, y′). Here, σ 2 is the process variance, and r(y, y′) is the correlation function, such as the Gaussian correlation function:

[0027] r(y, y′) = exp(δ(y - y′) 2 ) (8)

[0028] Among them, δ is the correlation parameter vector. Given the training set D = (Y, Z), the predicted value of the Kriging model at the unobserved point y satisfies the normal distribution:

[0029]

[0030] Among them, represents the predicted variable, and μ G (y) and σ G (y) represent the mean and standard deviation of the normal distribution followed respectively. The corresponding expressions are:

[0031] μ G (y) = β0 + r(y, Y)R -1 [Z - β0F] (10)

[0032]

[0033] Among them, r(y, Y) represents the correlation vector between the unobserved point x and the training set Y samples, and R -1 represents the inverse matrix of the autocorrelation matrix r(X, X) of the training set Y samples, and F represents the unit vector [1, 1,..., 1]T The superscript "T" represents transpose.

[0034] Setting 4: Assume that there are m performance indicators for the load-bearing structure material for which accelerated storage tests have been carried out. For a single stress indicator, the normal stress level is S0, and the stress levels are S1, S2..., S K , and the time measurement points are all t1, t2,...t n , corresponding to the time measurement point t i The measured degradation data of the jth sample is y i,j , j = 1, 2,..., r i . The storage period to be evaluated is [0, T], and the structural load-bearing threshold is w.

[0035] Based on the above settings, a method for evaluating the accelerated storage degradation of a load-bearing structure based on adaptive Kriging surrogate modeling proposed by the present invention mainly includes parameter estimation of the material-level accelerated degradation model, structural-level simulation test design and initial modeling, structural-level model adaptive iterative update and convergence discrimination, and structural-level load-bearing capacity degradation evaluation; the steps are as follows:

[0036] Step 1: Parameter estimation of the material-level accelerated degradation model

[0037] First, for the performance degradation test data of the material, respectively based on the linear degradation model, exponential degradation model, and power function degradation model, use the maximum likelihood estimation method to carry out parameter estimation of the accelerated degradation model. The likelihood functions are respectively:

[0038]

[0039] Among them, t i represents the ith test time measurement point, i = 1, 2..., n; S k represents the kth accelerated stress, k = 1, 2,..., K; y i,j represents the degradation data of the jth sample at the time measurement point t i , j = 1, 2,..., r i , r i represents the number of samples at the time measurement point t i ; θ(S k ) represents the acceleration model, which can be one of the Arrhenius model and the inverse power law model; y(t i,j |θ(S k )) represents the degradation model under the given acceleration model θ(S k ), which can be one of the linear degradation model, exponential degradation model, and power function degradation model.

[0040] The corresponding model parameter estimation values are:

[0041]

[0042] Among them, parameters characterizing the selected degradation model and acceleration model. The optimal model can be selected using the AIC criterion, that is:

[0043] AIC = 2τ - 2lnL (14)

[0044] where τ is the number of unknown parameters on the right side of equation (12). The model and parameter estimates that minimize the AIC value should be selected.

[0045] Step 2: Structural-level simulation test design and initial modeling

[0046] Use Latin hypercube sampling to conduct the initial test design of the structural simulation. First, generate a candidate set of standard normal random samples with dimension m and an initial sample size of 12, denoted as Uniformly and randomly generate a one-dimensional sample of the same size within the storage period [0, T], denoted as t0. By combining the evaluation results of the material-level parameters Give the input design parameters of the simulation test For the performance y ω , ω = 1, 2..., m, its design value can be calculated according to the selected model as:

[0047] Linear degradation model:

[0048]

[0049] Exponential degradation model:

[0050]

[0051] Power function degradation model:

[0052]

[0053] Substitute the initial sampling point Y0 into the simulation model G(y), and the obtained response value is denoted as Z0. Combining the above simulation input design parameters Y0, the initial training set can be expressed as D0 = (Y0, Z0).

[0054] Step 3: Adaptive iterative update and convergence discrimination of the structural-level model

[0055] Use Latin hypercube sampling to generate a candidate set U of standard normal random samples with size N LHS and dimension m LHS . Uniformly and randomly generate a one-dimensional row vector sample t with size N LHS within the storage period [0, T]. LHS Based on U LHS and t LHS, and the evaluation results of material-level parameters Similarly, substitute U LHS and t LHS into equations (15) to (17) to obtain the candidate sampling set Y LHS , that is

[0056] Linear degradation model:

[0057]

[0058] Exponential degradation model:

[0059]

[0060] Power function degradation model:

[0061]

[0062] where T LHS = [t LHS ; t LHS ;...; t LHS , and are the ω-th terms of Y LHS and U LHS respectively, and ω = 1, 2,..., m.

[0063] Substitute the initial training set D0 into equations (9) to (11) to obtain the predicted mean value LHS of the response at each sample point within the sampling set Y and its predicted variance The expressions are shown in equations (9) to (11), where y (p) ∈Y LHS , p = 1, 2,..., N LHS .

[0064] Try to determine the optimal sample point:

[0065]

[0066] Determine whether the following equation is satisfied:

[0067]

[0068] If it is satisfied, go to step four; if not, substitute the optimal sampling point y opt into the simulation model G(y), and record the obtained response value as z opt , and (y opt , z opt) Add it into the training set D0, rebuild the Kriging surrogate model based on the updated dataset, and restart from Step 3. We denote the converged training set as D c .

[0069] Step 4: Structural-level bearing capacity degradation reliability assessment

[0070] Given the storage period t, based on the sample set U LHS and the evaluation results of material-level parameters Similarly, substitute them into Eqs. (15) - (17) to obtain the candidate sampling set Y t , that is

[0071] Linear degradation model:

[0072]

[0073] Exponential degradation model:

[0074]

[0075] Power function degradation model:

[0076]

[0077] where t = [t; t;...; t], Y t ω and are the ω-th terms of Y t and U t respectively, ω = 1, 2,..., m.

[0078] Substitute the converged training set D c into Eqs. (9) - (11), and the predicted mean value t of the response at each sample point within the sampling set Y of the Kriging surrogate model and its predicted variance can be obtained. Their expressions are shown in Eqs. (9) - (11), where y (q) ∈Y t , q = 1, 2,..., N LHS . Therefore, the reliability at the storage period t can be evaluated as:

[0079]

[0080] where I(·) is the indicator function, which is 1 for positive values and 0 otherwise, and w is the failure threshold of the load-bearing structure.

[0081] The "Latin hypercube sampling" mentioned in Step 2 is a random sampling method that can be used for high-dimensional parameter spaces. It mainly divides the parameter space into layers to ensure uniform coverage of each parameter within its value range. Specifically, it can be achieved through the following steps:

[0082] a) Divide the value range of each parameter into uniform intervals.

[0083] b) Randomly select a value within each interval, ensuring that only one sample is taken from each interval.

[0084] c) Combine the values of all parameters to form sampling points.

[0085] The advantages and beneficial effects of the present invention are as follows:

[0086] ① The present invention is applicable to the accelerated test evaluation of load-bearing structure products;

[0087] ② The present invention solves the problem that traditional system integration methods cannot integrate structural physical information. It fuses physical structure characteristics through a surrogate model, and then transmits material-level accelerated degradation information to structural-level characteristics. Moreover, it has a low cost and broad application prospects. Description of the Drawings

[0088] Figure 1 is the flow chart of the method described in the present invention.

[0089] Figure 2 is the reliability curve within the storage period of 0 to 25 years. Detailed Embodiments

[0090] The present invention will be further described in detail below with reference to examples.

[0091] The cylindrical load-bearing structure of a special equipment carried out high-temperature accelerated degradation tests at temperature stress levels of 363.15K, 383.15K, 403.15K, and 423.15K respectively. A total of 3 performance indicators, namely A, B, and C, were measured. The accelerated degradation test data are shown in Tables 1 to 3.

[0092] Table 1 Accelerated Degradation Test Data of Performance Indicator A of a Load-Bearing Structure

[0093]

[0094] Table 3 Accelerated Degradation Test Data of Performance Indicator B of a Load-Bearing Structure

[0095]

[0096] Table 3 Accelerated Degradation Test Data of Performance Indicator C of a Load-Bearing Structure

[0097]

[0098] Assume that the simulation model of the load-bearing structure is known, and its simulation response characteristics can be obtained based on test parameters. Now, according to the method for accelerating the storage reliability assessment of the load-bearing structure based on adaptive Kriging surrogate modeling proposed in the present invention, the degradation modeling analysis of the load-bearing structure is carried out, and the storage reliability under a given storage period is predicted.

[0099] See Figure 1 , a method for accelerating the storage reliability assessment of a load-bearing structure based on adaptive Kriging surrogate modeling in the present invention is implemented through the following steps:

[0100] Step 1: Parameter estimation of the material-level accelerated degradation model

[0101] First, the degradation model parameters of performance indicators A, B, and C are estimated by maximum likelihood estimation, and according to the AIC criterion, the Arrhenius model is selected for the acceleration model. The parameter estimation results are:

[0102] Performance indicator A:

[0103] The optimal degradation model is a linear model, and the parameter estimation results are:

[0104]

[0105] The degradation model corresponding to the normal stress level S0 is:

[0106] y A = 3093 - 0.1305t + 88.11ε (28)

[0107] Performance indicator B:

[0108] The optimal degradation model is an exponential model, and the parameter estimation results are:

[0109]

[0110] The degradation model corresponding to the normal stress level S0 is:

[0111] y B = 51.3exp(-0.00449t) + 2.84ε (30)

[0112] Performance indicator C:

[0113] The optimal degradation model is an exponential model, and the parameter estimation results are:

[0114]

[0115] The degradation model corresponding to the normal stress level S0 is:

[0116] y C= 179.8exp(-0.003964t)+4.103ε (32)

[0117] Step 2: Structural-level simulation test design and initial modeling

[0118] Use Latin hypercube sampling to conduct the initial test design of structural simulation. First, generate a candidate set of standard normal random samples with dimension m and an initial sample size of 12, denoted as Uniformly and randomly generate one-dimensional samples of the same size within the storage period [0, 25] years, denoted as t0; by combining the evaluation results of material-level parameters Substitute U0 and t0 into equations (15) - (17), and the input design parameters of the simulation test can be obtained Based on this, obtain the simulation response value Z0. Therefore, the initial training set is D0 = (Y0, Z0), as shown in Table 4

[0119] Table 4 Accelerated degradation test data of the performance index C of a certain load-bearing structure

[0120]

[0121] Step 3: Adaptive iterative update and convergence discrimination of the structural-level model

[0122] Use Latin hypercube sampling to generate a candidate set U of standard normal random samples with size N LHS = 1000 and dimension m = 3 LHS . Uniformly and randomly generate a one-dimensional row vector sample t of size N LHS within the storage period [0, 25] years LHS . Based on U LHS and t LHS , and the evaluation results of material-level parameters Similarly, substitute U LHS and t LHS into equations (18) - (20) to obtain the candidate sampling set Y LHS . Based on the initial training set D0, obtain the predicted mean value LHS of the response at each sample point within the sampling set Y of the Kriging surrogate model and its predicted variance and perform adaptive iterative update based on equation (21). After 88 iterations of update, equation (22) converges, and the estimated results of the Kriging model parameters after convergence are as follows:

[0123] β0 = -0.0709, δ = [1.3243, 0.1962, 0.1, 0.1071] (33)

[0124] Step 4: Reliability assessment of the degradation of the structural load-bearing capacity

[0125] Given a storage period \(t = [0, 25]\) years, a sample set \(U\) with a sample size of \(N = 1000\) is regenerated. LHS And \(t = [t; t;...; t]\), similarly, the candidate sampling set \(Y\) can be obtained according to equations (23) - (25). Substitute \(Y\) into the converged Kriging surrogate model to obtain the response prediction value. Then, according to equation (26), the reliability at the storage period \(t = 25\) years is:

[0126] \(R(t)=0.381\)

[0127] The reliability curve within the storage period \(t = [0, 25]\) years is as Figure 2 shown.

[0128] In summary, the present invention provides an accelerated storage degradation assessment method for load-bearing structures based on adaptive Kriging surrogate modeling, that is, a storage reliability assessment method for load-bearing structures based on the Kriging surrogate model and the accelerated degradation assessment theory. It conducts degradation modeling on the mechanical properties at the material level of the load-bearing structure based on the accelerated degradation assessment theory, combines structural mechanics simulation response analysis and surrogate modeling theory, and further transfers the material-level degradation information to the structural level for carrying out structural-level degradation modeling and assessment of the load-bearing structure, solving the problems of low assessment accuracy and difficulty in conducting assessment in the traditional method for load-bearing structures. The specific steps of this method are: First, parameter estimation of the material-level accelerated degradation model; Second, structural-level simulation test design and initial modeling; Third, adaptive iterative update and convergence discrimination of the structural-level model; Fourth, structural-level load-bearing capacity degradation assessment. This method is applicable to the accelerated storage reliability assessment of load-bearing structures, without the need to conduct a large number of structural-level accelerated storage tests, and has the characteristics of high efficiency and low cost.

Claims

1. A method for accelerating storage degradation assessment of load-bearing structures based on adaptive Kriging proxy modeling requires the following basic settings: Setting 1: The storage degradation law of the structural material obeys one of the linear degradation, exponential degradation and power function degradation models; Setting 2: β, λ and η in the above three degradation models are acceleration related parameters; Setting 3: The simulation model of the structure is known, that is, when the m-dimensional performance parameter y is input, the response value of the structural bearing capacity can be obtained; Setting 4: Assume that there are m performance indicators of the load-bearing structural materials and an accelerated storage test is carried out; Based on the above configuration, it is characterized in that: The steps include: Step 1: Estimation of material-level accelerated degradation model parameters Firstly, the performance degradation test data of the material are used to estimate the parameters of the accelerated degradation model using the maximum likelihood estimation method based on the linear degradation model, exponential degradation model and power function degradation model. Step 2: Structural level simulation test design and initial modeling Latin hypercube sampling is used to carry out the initial experimental design of structural simulation. A standard normal random sample candidate set with dimension m and initial sample size of 12 is generated, which is denoted as Uniformly randomly generate one-dimensional samples of the same size within the storage period [0, T], denoted as t0; the results are evaluated by combining material-level parameters Given the input design parameters of the simulation test Step 3: Adaptive iterative update and convergence judgment of structural model Use Latin hypercube sampling to generate a size of N LHS , the standard normal random sample candidate set U with dimension m LHS ; Generate a random number of size N uniformly within the storage period [0,T] LHS The 1-dimensional row vector sample t LHS ; If the judgment condition is met, go to step 4; if not, the optimal sampling point y opt Substitute into the simulation model G(y), and the obtained response value is recorded as z opt , and (y opt ,z opt ) is added to the training set D0, the Kriging proxy model is rebuilt based on the updated data set, and step 3 is restarted; Step 4: Reliability assessment of structural load-bearing capacity degradation Given a storage period t, based on the sample set U LHS and material level parameter evaluation results The reliability at storage period t is evaluated.

2. The method for accelerating storage degradation assessment of load-bearing structures based on adaptive Kriging proxy modeling according to claim 1 is characterized in that: In setting 1, the expressions of each model are: ①Linear degradation model: y(t)=α+β·t+σ·ε (1) ②Exponential degradation model: y(t)=μ·exp(λ·t)+σ·ε (2) ③Power function degradation model: y(t)=η·t θ +s·e (3) Where t represents the storage time, α and β represent the linear degradation model parameters, μ and λ represent the exponential degradation model parameters, η and θ represent the power function degradation model parameters, ε represents a standard normal distribution variable with a mean of 0 and a variance of 1, and σ is the standard deviation.

3. The method for evaluating accelerated storage degradation of load-bearing structures based on adaptive Kriging proxy modeling according to claim 1 or 2, characterized in that: In setting 2, β, λ and η in the three degradation models are uniformly recorded as θ; the acceleration law of the structural material is considered to obey one of the Arrhenius model and the inverse power law model, and the corresponding model expressions are: Arrhenius model: lnθ=a+b / S (4) Inverse power law model: lnθ=a+blnS (5) Where θ represents the acceleration-related parameter in the degradation model, a and b represent the acceleration model parameters, and S represents the acceleration stress.

4. The method for accelerating storage degradation assessment of load-bearing structures based on adaptive Kriging proxy modeling according to claim 1 is characterized in that: In setting 3, the input-output relationship is defined as: z=G(y) (6) Among them, z represents the structural bearing capacity response value, G(y) represents the response model; The Kriging model is used to replace the above input-output relationship, namely: G(y)=β0+z G (y) (7) Among them, β0 is a constant that characterizes the global trend, z G (y) has zero mean and covariance σ 2 r(y,y′) is a stationary Gaussian process; here, σ 2 is the process variance, r(y,y′) is the correlation function, such as the Gaussian correlation function: r(y,y′)=exp(δ(yy′) 2 ) (8) Where δ is the relevant parameter vector; given the training set D = (Y, Z), the estimated value of the Kriging model at the unobserved point y satisfies the normal distribution: in, represents the estimated variable, μ G (y) and σ G (y) represent the mean and standard deviation of the normal distribution respectively; the corresponding expressions are: μ G (y)=β0+r(y,Y)R -1 [Z-β0F] (10) Among them, r(y,Y) represents the correlation vector between the unobserved point x and the training set Y sample, R -1 represents the inverse matrix of the autocorrelation matrix r(X,X) of the training set Y samples, and F represents the unit vector [1,1,...,1] T , the superscript "T" indicates transpose.

5. The method for accelerating storage degradation assessment of load-bearing structures based on adaptive Kriging proxy modeling according to claim 1 is characterized in that: In setting 4, for a single stress indicator, the normal stress level is S0, and the stress levels are S1, S2…, S K The time measurement points are t1, t2, …t n , corresponding to the time measurement point t i The measured degradation data of the jth sample is y i,j ,j=1,2,...,r i ; The storage period to be evaluated is [0,T], and the structural load threshold is w.

6. The method for accelerating storage degradation assessment of load-bearing structures based on adaptive Kriging proxy modeling according to claim 1 is characterized in that: In step 1, the likelihood functions are: Among them, t i represents the i-th test time measurement point, i=1,2...,n; S k represents the kth acceleration stress, k = 1, 2, ..., K; y i,j Indicates the time measurement point t i The jth sample degradation data under i , r i Indicates the time measurement point t i The number of samples under θ(S k ) represents the acceleration model, which adopts one of the Arrhenius model and the inverse power law model; y(t i,j |θ(S k )) Given the acceleration model θ(S k ) is a degradation model under the condition of adopting one of a linear degradation model, an exponential degradation model and a power function degradation model; The corresponding model parameter estimates are: in, Characterize the parameters of the selected degradation model and acceleration model; use the AIC criterion to select the optimal model, that is: AIC=2τ-2lnL (14) Where τ is the number of unknown parameters on the right side of equation (12); the model and parameter estimates that minimize the AIC value should be selected.

7. The method for accelerating storage degradation assessment of load-bearing structures based on adaptive Kriging proxy modeling according to claim 1 is characterized in that: In step 2, the design value is calculated based on the selected model as: Linear degradation model: Exponential degradation model: Power function degradation model: Substitute the initial sampling point Y0 into the simulation model G(y), and the obtained response value is recorded as Z0; combined with the above simulation input design parameter Y0, the initial training set is expressed as D0 = (Y0, Z0).

8. The method for evaluating accelerated storage degradation of load-bearing structures based on adaptive Kriging proxy modeling according to claim 6 or 7, characterized in that: Latin hypercube sampling is implemented by the following steps: a) Divide the value range of each parameter into uniform intervals; b) Randomly select a value in each interval, ensuring that only one sample is taken in each interval; c) Combine the values ​​of all parameters to form sampling points.

9. The method for accelerating storage degradation assessment of load-bearing structures based on adaptive Kriging proxy modeling according to claim 7 is characterized in that: In step three, based on U LHS and t LHS And material level parameter evaluation results Will U LHS and t LHS Substitute into equations (15) to (17) to obtain the candidate sampling set Y LHS ,Right now Linear degradation model: Exponential degradation model: Power function degradation model: Among them, T LHS =[t LHS ;t LHS ;…;t LHS ], and Y LHS and U LHS The ωth term of , ω=1,2,…,m; Get the Kriging surrogate model in the sampling set Y LHS The estimated mean of the response at each sample point in and its estimated variance where y (p) ∈Y LHS ,p=1,2,...,N LHS ; Try to determine the optimal sample point: Determine whether the following equation is satisfied:

10. The method for accelerating storage degradation assessment of load-bearing structures based on adaptive Kriging proxy modeling according to claim 1, characterized in that: In step 4, obtain the candidate sampling set Y t ,include: Linear degradation model: Exponential degradation model: Power function degradation model: where t = [t; t; ...; t], and Y t and U t The ωth term of , ω=1,2,...,m; Get the Kriging surrogate model in the sampling set Y t The estimated mean of the response at each sample point in and its estimated variance where y (q) ∈Y t , q=1,2,...,N LHS ; Therefore, the reliability at storage period t can be evaluated as: Where I(·) is the indicator function, which is 1 if positive and 0 otherwise, and w is the failure threshold of the load-bearing structure.

Images