Ship beam strength reliability detection method and system based on learning function

Abstract

The invention relates to the technical field of ship structure analysis, in particular to a hull beam strength reliability detection method and system based on a learning function, and the method comprises the steps: obtaining variable data influencing the ultimate strength of a hull beam, and determining a probability density function and a hull beam ultimate strength performance function according to the variable data; constructing a performance function agent model based on a Kriging model, and obtaining a response value set of the candidate sample set according to the performance function agent model, including obtaining the candidate sample set close to a limit response surface by using importance sampling and a subset simulation framework, judging the convergence of the performance function agent model by using a learning function, and obtaining a response value set of the candidate sample set according to the response value set. According to the method, the candidate sample set close to the limit response surface is obtained by adopting the importance sampling and subset simulation framework, the importance sampling is used for constructing a proper sampling density function, and the subset simulation is used for introducing an intermediate failure event, so that the samples near the limit response surface are effectively captured, and the accuracy of failure probability calculation is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ship structure analysis, and in particular to a hull beam strength reliability detection method and system based on a learning function. Background Art

[0002] Against the backdrop of the booming global shipping industry and the gradual development of the Arctic route, the sailing areas of ships are constantly expanding, and the probability of encountering extreme sea conditions has increased significantly. The marine environment is complex and changeable. Under harsh conditions, the hull beams of ships are always under tremendous pressure, which poses an extremely severe challenge to their ultimate strength.

[0003] The ultimate strength of hull beam is a key indicator to measure the ability of ship structure to resist damage under extreme wave loads, and its reliability assessment is of vital importance. Traditional hull beam ultimate strength assessment mostly adopts deterministic methods, which has obvious defects. It often ignores the uncertainty of many factors such as wave loads, hull structure, and material properties. However, in actual situations, the uncertainty of these factors has a significant impact on the ultimate strength of hull beam. In addition, in terms of evaluating the reliability of ship structures, the commonly used methods are mainly Monte Carlo sampling (MCS). The MCS method has been widely used due to its advantages such as simple operation and high accuracy. However, this method has obvious shortcomings when dealing with low-probability failure events. In order to obtain accurate results, a large number of function calls are required for simulation sampling.

[0004] Finally, patent number "2020106372802" uses low-discrepancy sequence sampling methods such as Latin hypercube sampling to generate candidate sample points. Although the uniform distribution of candidate sample points in the sampling space is guaranteed, as the dimension of the structural functional function increases, the sample space expands exponentially. Even if low-discrepancy sequence sampling is used, it becomes increasingly difficult to maintain uniform distribution of samples in high-dimensional space and effectively cover all important areas. The number of sample points needs to be greatly increased to maintain the same coverage accuracy. The amount of calculation and storage requirements increase dramatically, and the "curse of dimensionality" problem is prone to occur, causing the sample points to become sparse in high-dimensional space, unable to fully reflect the complex relationship between variables, and reducing the effectiveness of high-dimensional structural reliability analysis. At this stage, a hull beam strength reliability detection method and system based on learning functions is needed. Summary of the Invention

[0005] In order to solve the problem of low accuracy of hull girder strength detection due to ignoring the influence of uncertain factors and incomplete evaluation sampling, the present invention provides a hull girder strength reliability detection method and system based on learning function.

[0006] In a first aspect, the present invention provides a hull girder strength reliability detection method based on a learning function, which adopts the following technical solutions:

[0007] A hull girder strength reliability detection method based on a learning function, comprising:

[0008] S1. Obtain variable data that affects the ultimate strength of the hull girder, and determine a probability density function and a hull girder ultimate strength function function based on the variable data;

[0009] S2. constructing a functional function proxy model based on the Kriging model, including using the Kriging model to construct a mapping relationship between variable data and functional function output values;

[0010] S3. Obtaining a set of response values ​​of a candidate sample set according to a performance function surrogate model, including obtaining a candidate sample set close to a limit response surface using importance sampling and a subset simulation framework;

[0011] S4, using the learning function to judge the convergence of the performance function proxy model, including calculating the learning function value of each sample point in the candidate sample set;

[0012] S5. When the judgment result indicates non-convergence, adaptively updating the hull girder strength assessment model, including selecting the best sample point according to the learning function value, and updating the model according to the best sample point;

[0013] S6. When the judgment result indicates convergence, the performance function proxy model is used to obtain and verify the final failure probability.

[0014] Furthermore, the method of determining the probability density function and the ultimate strength function of the hull girder according to the variable data includes obtaining distribution information of the random variable as the probability density function, and constructing the ultimate strength function of the hull girder according to the random variable. The function expression is:

[0015]

[0016] Among them, M u is the ultimate longitudinal bending moment of the hull girder, M S and M w are the still water bending moment and wave bending moment, respectively, u , χ S , χ w M u 、M S and M w The model uncertainty parameter, Load reduction factor applied to account for the superposition of SWBM and VWBM.

[0017] Furthermore, the construction of the functional function proxy model based on the Kriging model includes the following steps:

[0018] S21. Use the Halton sampling method to extract multiple sample points in the sampling space of the variable, and form the multiple sample points into an initial sample set;

[0019] S22, obtaining the initial response value of the hull girder ultimate strength function corresponding to the initial sample set;

[0020] S23, construct a functional function proxy model using the Kriging model based on the initial sample set and initial response value

[0021] Furthermore, the method of using importance sampling and subset simulation framework to obtain a candidate sample set close to the limit response surface includes: S31, setting an initial value of the number of subset simulation layers P, and extracting multiple sample points in the sampling space based on the Monte Carlo sampling method to form an initial candidate sample set Ω;

[0022] S32, Utilization Get the response value Y of the candidate sample set SS , sort the response values ​​in descending order, and select the response value according to the sequence as the critical value of the subset simulation intermediate event.

[0023] S33. When the critical value is less than or equal to 0, the failure probability of the hull girder is calculated using the indicator function, and the entire process ends. When the critical value is greater than 0, the next step is performed.

[0024] S34, update the number of subset simulation layers P according to the step size of 1, and record in, represents the failure probability, and P represents the number of subset simulation layers;

[0025] S35, constructing an importance sampling density function according to the critical value, and performing Markov chain Monte Carlo sampling based on the importance sampling density function to obtain an updated candidate point set Ω, and transferring the updated candidate point set to step S32 to repeat the above operation;

[0026] S36. Output candidate sample set Ω and response value Y SS and hull girder failure probability

[0027] Furthermore, the Monte Carlo sampling method is used to extract multiple sample points in the sampling space to form an initial candidate sample set, including constructing the sampling density function required by Markov chain Monte Carlo based on the importance sampling method, and obtaining the probability density function q(x|F i-1 ) candidate sample set Ω, and define the limit distribution of the Markov chain as F i-1 The conditional probability density distribution of the region, the conditional probability density distribution formula is:

[0028]

[0029] in, Expressed as an indicator function, when the sample point x belongs to F i-1 hour, When it doesn't belong, f X (x) represents the proposed distribution for controlling the state transition in the Markov chain process, P(F i-1 ) is event F i-1 probability of failure.

[0030] Furthermore, the convergence of the functional function proxy model is judged by using the learning function, including calculating the learning function value for each sample point in the candidate sample set Ω by using the learning function, setting a convergence threshold, and calculating the convergence according to the learning function value and the convergence threshold. The convergence calculation formula is:

[0031]

[0032] Among them, ER(x i ) represents the kriging model of the kth iteration At point u i The learning function value of Expressed as a kriging model At point u i The mean square error of the prediction, n M Indicates the total number of sample points.

[0033] Furthermore, when the judgment result indicates that the model has not converged, the hull beam strength evaluation model is adaptively updated, including selecting the sample point with the largest learning function value in the candidate sample set Ω as the best sample x according to the learning function value. new , add the best sample to the initial sample set, and use the initial sample set after adding the best sample to rebuild the functional function agent model

[0034] In a second aspect, a hull girder strength reliability detection system based on a learning function includes:

[0035] The data acquisition module is configured to: acquire variable data affecting the ultimate strength of the hull beam, and determine a probability density function and a function function of the ultimate strength of the hull beam based on the variable data;

[0036] The model module is configured to: construct a functional function proxy model based on the Kriging model, including using the Kriging model to construct a mapping relationship between variable data and functional function output values;

[0037] The set module is configured to: obtain a response value set of a candidate sample set according to a performance function proxy model, including obtaining a candidate sample set close to a limit response surface by using importance sampling and a subset simulation framework;

[0038] The convergence module is configured to: use the learning function to judge the convergence of the performance function proxy model, including calculating the learning function value of each sample point in the candidate sample set;

[0039] An updating module is configured to: when a judgment result indicates non-convergence, adaptively update the hull girder strength evaluation model, including selecting an optimal sample point according to a learning function value, and updating the model according to the optimal sample point;

[0040] The output module is configured to: when the judgment result indicates convergence, use the functional function proxy model to obtain and verify the final failure probability.

[0041] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for example, a method for detecting the strength reliability of a hull beam based on a learning function.

[0042] In a fourth aspect, the present invention provides a terminal device comprising a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor to implement the method for detecting the strength reliability of a hull beam based on a learning function.

[0043] In summary, the present invention has the following beneficial technical effects:

[0044] 1. The present invention uses Halton sampling and Kriging model to construct a performance function proxy model. Halton sampling can efficiently extract representative sample points in the variable sampling space and reduce sample redundancy. The Kriging model, with its good local and global approximation capabilities, constructs an accurate mapping relationship between variable data and performance function output values ​​based on these sample points. Compared with directly calculating complex performance functions, the proxy model greatly reduces the computing cost and improves the computing efficiency, making it more efficient and feasible when processing large-scale computing tasks.

[0045] 2. The present invention adopts importance sampling and subset simulation framework to obtain candidate sample sets close to the limit response surface. Importance sampling guides the sampling process to focus on areas that have an important impact on the failure probability by constructing a suitable sampling density function, increases the probability of samples falling into the failure area, and reduces invalid sampling. Subset simulation introduces intermediate failure events, decomposes complex reliability problems into multiple simple conditional probability problems, and gradually approaches the limit response surface. The combination of the two can effectively capture samples near the limit response surface, improve the ability to capture low-probability failure events, and thus improve the accuracy of failure probability calculation.

[0046] 3. The present invention uses a learning function to judge the convergence of the functional function proxy model. By calculating the learning function value of each sample point in the candidate sample set and calculating the convergence index according to a specific formula, and comparing it with the preset convergence threshold, it can accurately judge whether the model has converged. The adaptive update mechanism ensures that the model can be continuously optimized, gradually approaches the actual situation as the number of iterations increases, and improves the fitting accuracy and reliability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of a hull girder ultimate strength reliability analysis method combined with a novel learning function provided by the present invention;

[0048] Figure 2 1 is a schematic diagram showing the final sampling result of the method proposed in the present invention and the fitting of the kriging model to the true limit state function of the structure in Example 1 of the present invention;

[0049] Figure 3 Schematic diagram showing the final sampling results of the improved typical method in Example 1 of the present invention and the fitting of the kriging model to the true limit state function of the structure;

[0050] Figure 4 3 is a schematic diagram of the convergence process of the adaptive structural reliability analysis and failure probability prediction method of the present invention in Example 1 of the present invention. DETAILED DESCRIPTION

[0051] The present invention will be further described in detail below with reference to the accompanying drawings.

[0052] Example 1

[0053] Reference Figure 1 A hull girder strength reliability detection method based on a learning function in this embodiment includes:

[0054] S1. Obtain variable data that affects the ultimate strength of the hull girder, and determine a probability density function and a hull girder ultimate strength function function based on the variable data;

[0055] S2. constructing a functional function proxy model based on the Kriging model, including using the Kriging model to construct a mapping relationship between variable data and functional function output values;

[0056] S3. Obtaining a set of response values ​​of a candidate sample set according to a performance function surrogate model, including obtaining a candidate sample set close to a limit response surface using importance sampling and a subset simulation framework;

[0057] S4, using the learning function to judge the convergence of the performance function proxy model, including calculating the learning function value of each sample point in the candidate sample set;

[0058] S5. When the judgment result indicates non-convergence, adaptively updating the hull girder strength assessment model, including selecting the best sample point according to the learning function value, and updating the model according to the best sample point;

[0059] S6. When the judgment result indicates convergence, the performance function proxy model is used to obtain and verify the final failure probability.

[0060] Specifically;

[0061] S1. Obtain variable data that affects the ultimate strength of the hull girder, and determine a probability density function and a hull girder ultimate strength function function based on the variable data;

[0062] like Figure 1 As shown in the figure, when evaluating the reliability of the ultimate strength of the hull girder, the random variables that affect the failure probability of the ultimate strength of the hull girder are first determined. These random variables usually include structural dimensions (such as plate thickness, frame size, etc.), material properties (such as yield strength, elastic modulus, etc.), wave bending moment and model uncertainty parameters. In this embodiment, the random variables are represented by X = [χ u ,χ S ,χ w ,M s ,M w ,E,σ Yep ,t p ,t w ,t f ] indicates that for the above random variables, the distribution information f(x) is obtained. The distribution information describes the probability law of the random variable value, that is, the probability density function f(x). Different random variables may have different distribution types, such as normal distribution, lognormal distribution, uniform distribution, etc. In this embodiment, the random variable obeys the standard normal distribution. The function g(x) is a mathematical expression used to describe the ultimate strength state of the hull beam. When g(x)>0, it means that the hull beam is in a safe state. When g(x)≤0, it means that the hull beam is in a failed state. According to the random variable, the function function g(x) is defined as:

[0063]

[0064] Among them, M u is the ultimate longitudinal bending moment of the hull girder, M S and M w are the still water bending moment and wave bending moment, respectively, u , χ S , χ w M u 、M S and M w The model uncertainty parameter, The load reduction factor applied to consider the superposition of SWBM and VWBM is Elastic modulus E, yield stress χ Yep , and the plate thickness t of the strip plate, stiffener web plate and stiffener flange plate o , t w and t f is a random variable, the wave bending moment is a random variable elastic modulus E, material yield stress σ Yep , and the thickness of the strip plate, stiffener web plate and stiffener flange plate t p , t w and t f function, namely M u ~f(E,σ Yep ,t p ,t w ,t f ), through this function, the value of the random variable can be linked to the ultimate strength state of the hull beam, so as to perform reliability evaluation.

[0065] S2. constructing a functional function proxy model based on the Kriging model, including using the Kriging model to construct a mapping relationship between variable data and functional function output values;

[0066] like Figure 2 、 Figure 3 As shown in the figure, this step aims to construct a mapping relationship between variables and performance function output values ​​with the help of the Kriging model. The core of this step is to obtain sample points by sampling, calculate the performance function response values ​​corresponding to the sample points, and then use these data to construct the Kriging proxy model. The space R is determined according to the probability density function f(x) of the variable determined in step 1. The sampling space can be expressed as:

[0067] in, Expressed as the joint probability density function f of the variables x(x), Φ is the cumulative distribution function of the standard normal distribution. Since Φ(±6) covers almost all possible values ​​in the standard normal distribution, the sampling space R can cover most of the possible values ​​of the variable. The Halton sampling method is used to extract N in the determined sampling space R. H = 15 sample points, the N H The sample points are combined into the initial sample set ξ. The initial sample set ξ is actually a set of random variable combinations in the sampling space R. Each sample point represents the value of a set of random variables. Halton sampling is a low-discrepancy sequence sampling method that can evenly distribute sample points in the sampling space. Compared with random sampling, it can more effectively cover the sampling space, thereby reducing the redundancy of sample points and improving the efficiency of subsequent calculations.

[0068] According to the performance function g(X) obtained in step 1, each sample point in the initial sample set ξ is substituted into the performance function g(X) for calculation, thereby obtaining the performance function response value Y corresponding to each sample point ξ , and then use the Kriging model to build a proxy model of the functional function. The Kriging model is a statistical interpolation method that assumes that there is a certain spatial correlation between variables and predicts the function value of the unknown point through the information of the known sample points. The basic form of the Kriging model can be expressed as:

[0069] G(x)=μ+Z(x),

[0070] Among them, μ is the global mean, Z(x) represents a zero-mean random process, based on the sample set ξ and the corresponding performance function response value Y ξ , use the Kriging model to build a proxy model of the current function Specifically, by fitting the sample points and the corresponding response values, the parameters in the Kriging model (such as the global mean, the parameters of the covariance function, etc.) are determined, thereby obtaining a proxy model that can approximate the performance function g(X) The functional agent model is expressed as:

[0071]

[0072] Among them, f T (x) is the basis function vector, β is the regression coefficient to be solved, which is solved using the initial sample set and the corresponding response value, and z(x) is a steady-state Gaussian process.

[0073] S3. Obtaining a set of response values ​​of a candidate sample set according to a performance function surrogate model, including obtaining a candidate sample set close to a limit response surface using importance sampling and a subset simulation framework;

[0074] (3.1) Initialize subset simulation and extract initial candidate sample set

[0075] Set the number of subset simulation layers P to 1, and gradually increase the number of simulation layers to make the sample approach the limit response surface. Then, use the Monte Carlo sampling method to extract n sample points in the variable sampling space R determined in step 1. These sample points can represent various possible situations of the variable within its value range. Combine the extracted n sample points into the initial candidate sample set Ω.

[0076] (3.2) Bring the candidate sample x in the candidate sample set Ω into the proxy model constructed in step 2 Get the response value Y SS , the response value Y of the candidate sample set Ω SS Arrange in descending order from largest to smallest, and record the response value Y at the 0.9n+1 position after sorting. 0.9 n +1 , using it as a subset to simulate the intermediate event F i ={x:g(x)≤b i The intermediate events here are used to gradually narrow the sample space and make the sample closer to the limit response surface.

[0077] (3.3) Determine the critical value and calculate the failure probability

[0078] Determine the critical value Y 0.9n+1 Is it less than or equal to 0? If Y (0.9n+1) ≤0, indicating that the current sample set is close enough to the limit response surface, and the final failure probability can be calculated. The calculation of the failure probability is specifically as follows:

[0079]

[0080] Among them, I(Y SS ) is the indicator function, when Y SS ≤0|, I(Y SS )=1; when Y SS >0, I(Y SS )=0. ∑I(Y SS ) represents the number of sample points whose response value is less than or equal to 0 in the sample set. Indicates the proportion of failed samples in the current sample set, P is the number of subset simulation layers, 0.1 P-1 Is a coefficient related to the number of subset simulation layers, which is used to adjust the estimated value of the failure probability. After the calculation is completed, go to step (3.6) to end the process. If Y (0.9n+1) >0, it means that the current sample set is not close enough to the limit response surface, and it is necessary to continue the simulation and go to step (3.4).

[0081] (3.4) Update the number of subset simulation layers and failure probability estimation

[0082] Update the number of subset simulation layers P to P+1 and record As the number of simulation layers increases, It will continue to adjust and gradually approach the actual failure probability.

[0083] (3.5) According to the critical value Y (0.9n+1) Construct an importance sampling density function. The purpose of importance sampling is to concentrate sampling more closely on regions that have a significant impact on the failure probability calculation, i.e., regions close to the limit response surface. This critical region can be determined by the critical value, thereby constructing the corresponding sampling density function. Markov chain Monte Carlo (MCMC) sampling is performed based on the constructed importance sampling density function to obtain an updated candidate point set Ω. MCMC sampling is a random sampling method based on a Markov chain that can generate samples under a given sampling density function. After sampling is completed, go to step (3.2) to recalculate the response value of the candidate sample set and continue the simulation iteration.

[0084] (3.6) When the critical value iY (0.9n+1) When ≤0, the importance sampling + subset simulation process ends and the candidate sample set Ω and response value Y are output. SS and hull girder failure probability

[0085] S4, using the learning function to judge the convergence of the performance function proxy model, including calculating the learning function value of each sample point in the candidate sample set;

[0086] like Figure 4 As shown, for each sample point x in the candidate sample set Ω i (i=1,2,…n M , n M Represents the total number of sample points), calculate the Kriging model of the kth iteration The learning function value ER(x i ), which reflects the prediction performance of the model at the sample point and is related to the prediction error or uncertainty. The prediction mean square error is then calculated. The larger the mean square error, the more unstable the prediction result of the model at that point and the lower the credibility. On the contrary, it means that the prediction result is more reliable. During the construction of the Kriging model, the mean square error estimate of the prediction value of each point can be obtained by learning the sample points. It is calculated based on the covariance function and parameter estimation of the model. Finally, the convergence threshold is set. In this embodiment, the convergence threshold is κ = 1×10 -4 , calculate the convergence index and compare it with the set threshold. The convergence calculation formula is:

[0087]

[0088] Among them, ER(x i ) represents the kriging model of the kth iteration At point u i The learning function value of Expressed as a kriging model At point u i The mean square error of the prediction, n M Represents the total number of sample points. If W is less than or equal to the threshold, it means that the Kriging model The model has achieved sufficient accuracy under the current sample set and can approximate the functional function relatively accurately. It can be considered that the model has converged and can proceed to the next step of using the model to calculate and verify the failure probability. If W is greater than the threshold, it indicates that the model prediction error or uncertainty is high and has not yet fully converged. The model needs to be updated and optimized until it meets the convergence requirements.

[0089] S5. When the judgment result indicates non-convergence, adaptively updating the hull girder strength assessment model, including selecting the best sample point according to the learning function value, and updating the model according to the best sample point;

[0090] If W is greater than the convergence threshold, it indicates that the prediction error or uncertainty of the model under the current sample set is high and has not yet reached the ideal accuracy. At this time, step S5 is entered to adaptively update the model to optimize the model performance so that it can better approach the real function. The learning function is used to select the best sample x in the candidate sample set Ω. new , the learning function expression is:

[0091]

[0092] in, Expressed as Kriging model The mean square error of the prediction at point x reflects the uncertainty of the model's prediction value at that point. It is the Kriging model The predicted mean at point x represents the model's predicted estimate of the function value at that point, and Φ is the cumulative distribution function of the standard normal distribution. is the mean square error Perform logarithmic transformation and add a constant term Adjust its influence weight on the learning function value, in the candidate sample set Ω, through the formula The meaning of this formula is to find the sample point in the candidate sample set Ω that maximizes the learning function ER(x). Because the larger the ER(x) value, the higher the prediction uncertainty of the model at that sample point.

[0093] The best sample point xnew Add to the initial sample set, the sample set now contains more valuable information for model construction. Based on the new sample set after adding the best sample, re-execute step S2, that is, use the Kriging model to build a new functional function proxy model Because the newly constructed model incorporates information about the optimal sample points, it can better fit the performance function, reduce prediction errors, and gradually approach a convergent state. By continuously repeating this adaptive update process (continuously selecting the optimal sample points, updating the sample set, and rebuilding the model when convergence is not achieved), the Kriging model increasingly accurately approximates the true performance function, providing a more reliable basis for hull girder strength reliability assessment.

[0094] S6. When the judgment result indicates convergence, the performance function proxy model is used to obtain and verify the final failure probability.

[0095] If the convergence index W is less than or equal to the set convergence threshold, it indicates that the Kriging model The sample data has been fully learned and can accurately approximate the real function. At this time, the model has reached a convergence state and can enter the final failure probability calculation and verification stage. Based on the converged Kriging model Extract a large number of sample points in the sampling space of the variable. Substitute these sample points into the Kriging model Calculate the response value for each sample point. Based on the performance function rule for determining the hull girder's condition (a hull girder failure occurs when the performance function value is less than or equal to 0), count the number of sample points with a response value less than or equal to 0. Estimate the failure probability of the hull girder by taking the ratio of the number of sample points with a response value less than or equal to 0 to the total number of samples.

[0096] Example 2

[0097] This embodiment differs from the first embodiment in that this embodiment provides a hull girder strength reliability detection system based on a learning function, including:

[0098] The data acquisition module is configured to: acquire variable data affecting the ultimate strength of the hull beam, and determine a probability density function and a function function of the ultimate strength of the hull beam based on the variable data;

[0099] The model module is configured to: construct a functional function proxy model based on the Kriging model, including using the Kriging model to construct a mapping relationship between variable data and functional function output values;

[0100] The set module is configured to: obtain a response value set of a candidate sample set according to a performance function proxy model, including obtaining a candidate sample set close to a limit response surface by using importance sampling and a subset simulation framework;

[0101] The convergence module is configured to: use the learning function to judge the convergence of the performance function proxy model, including calculating the learning function value of each sample point in the candidate sample set;

[0102] An updating module is configured to: when a judgment result indicates non-convergence, adaptively update the hull girder strength evaluation model, including selecting an optimal sample point according to a learning function value, and updating the model according to the optimal sample point;

[0103] The output module is configured to: when the judgment result indicates convergence, use the functional function proxy model to obtain and verify the final failure probability.

[0104] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, a method for detecting the strength reliability of a hull beam based on a learning function.

[0105] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor for a hull beam strength reliability detection method based on a learning function.

[0106] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A hull girder strength reliability detection method based on learning function, characterized in that: include: S1. Obtain variable data that affects the ultimate strength of the hull girder, and determine a probability density function and a hull girder ultimate strength function function based on the variable data; S2. constructing a functional function proxy model based on the Kriging model, including using the Kriging model to construct a mapping relationship between variable data and functional function output values; S3. Obtaining a set of response values ​​of a candidate sample set according to a performance function surrogate model, including obtaining a candidate sample set close to a limit response surface using importance sampling and a subset simulation framework; S4, using the learning function to judge the convergence of the performance function proxy model, including calculating the learning function value of each sample point in the candidate sample set; S5. When the judgment result indicates non-convergence, adaptively updating the hull girder strength assessment model, including selecting the best sample point according to the learning function value, and updating the model according to the best sample point; S6. When the judgment result indicates convergence, the performance function proxy model is used to obtain and verify the final failure probability.

2. A hull girder strength reliability detection method based on learning function according to claim 1, characterized in that: The method of determining the probability density function and the ultimate strength function of the hull beam according to the variable data includes obtaining the distribution information of the random variable as the probability density function, and constructing the ultimate strength function of the hull beam according to the random variable. The expression of the function is: Among them, M u is the ultimate longitudinal bending moment of the hull girder, M S and M w are the still water bending moment and wave bending moment, respectively, u , χ S , χ w M u 、M S and M w The model uncertainty parameter, Load reduction factor applied to account for the superposition of still water bending moment and vertical wave bending moment.

3. The method for detecting the strength reliability of a hull beam based on a learning function according to claim 1, characterized in that: The method of constructing a functional function proxy model based on the Kriging model includes the following steps: S21. Use the Halton sampling method to extract multiple sample points in the sampling space of the variable, and form the multiple sample points into an initial sample set; S22, obtaining the initial response value of the hull girder ultimate strength function corresponding to the initial sample set; S23, construct a functional function proxy model using the Kriging model based on the initial sample set and initial response value The functional agent model is expressed as: Among them, f T (x) is the basis function vector, β is the regression coefficient to be solved, which is solved using the initial sample set and the corresponding response value, and z(x) is a steady-state Gaussian process.

4. The method for detecting the reliability of hull beam strength based on learning function according to claim 1, characterized in that: The method of using importance sampling and subset simulation framework to obtain a candidate sample set close to the limit response surface includes: S31, setting an initial value of the number of subset simulation layers P, and extracting multiple sample points in the sampling space based on the Monte Carlo sampling method to form an initial candidate sample set Ω; S32. Bring the candidate sample x in the candidate sample set Ω into the model Get the response value Y SS , and sort the response values ​​in descending order, and select the response value Y according to the sequence 0.9n+1 As the critical value of the intermediate event of the subset simulation, where n is the total number of candidate samples x; S33. When the critical value is less than or equal to 0, the failure probability of the hull girder is calculated using the indicator function, and S36 is executed to end the entire process. When the critical value is greater than 0, the next operation is performed; S34, update the number of subset simulation layers P according to the step size of 1, and record in, represents the failure probability, P represents the number of subset simulation layers; S35, constructing an importance sampling density function according to the critical value, and performing Markov chain Monte Carlo sampling based on the importance sampling density function to obtain an updated candidate point set Ω, and transferring the updated candidate point set to step S32 to repeat the above operation; S36. Output candidate sample set Ω and response value Y SS and hull girder failure probability 5. A hull girder strength reliability detection method based on learning function according to claim 4, characterized in that: The Monte Carlo sampling method is based on extracting multiple sample points in the sampling space to form an initial candidate sample set, including constructing the sampling density function required by Markov chain Monte Carlo based on the importance sampling method, and obtaining the probability density function q(x|F i-1 ) candidate sample set Ω, and define the limit distribution of the Markov chain as F i-1 The conditional probability density distribution of the region, the conditional probability density distribution formula is: in, Expressed as an indicator function, when the sample point x belongs to F i-1 hour, When it doesn't belong, 0,f X (x) represents the proposed distribution for controlling the state transition in the Markov chain process, P(F i-1 ) is event F i-1 probability of failure.

6. A hull girder strength reliability detection method based on learning function according to claim 1, characterized in that: The convergence of the functional function proxy model is judged by using the learning function, including calculating the learning function value for each sample point in the candidate sample set Ω by using the learning function, setting a convergence threshold, and calculating the convergence according to the learning function value and the convergence threshold. The convergence calculation formula is: Among them, ER(x i ) represents the kriging model of the kth iteration At point u i The learning function value of Expressed as a kriging model At point u i The mean square error of the prediction, n M Indicates the total number of sample points.

7. A hull girder strength reliability detection method based on learning function according to claim 6, characterized in that: When the judgment result indicates that the model has not converged, the hull beam strength evaluation model is adaptively updated, including selecting the sample point with the largest learning function value in the candidate sample set Ω as the best sample x according to the learning function value. new , add the best sample to the initial sample set, and use the initial sample set after adding the best sample to rebuild the functional function agent model 8. A hull girder strength reliability detection system based on a learning function, executing the method according to any one of claims 1 to 7, characterized in that: include: The data acquisition module is configured to: acquire variable data affecting the ultimate strength of the hull beam, and determine a probability density function and a function function of the ultimate strength of the hull beam based on the variable data; The model module is configured to: construct a functional function proxy model based on the Kriging model, including using the Kriging model to construct a mapping relationship between variable data and functional function output values; The set module is configured to: obtain a response value set of a candidate sample set according to a performance function proxy model, including obtaining a candidate sample set close to a limit response surface by using importance sampling and a subset simulation framework; The convergence module is configured to: use the learning function to judge the convergence of the performance function proxy model, including calculating the learning function value of each sample point in the candidate sample set; An updating module is configured to: when a judgment result indicates non-convergence, adaptively update the hull girder strength evaluation model, including selecting an optimal sample point according to a learning function value, and updating the model according to the optimal sample point; The output module is configured to: when the judgment result indicates convergence, use the functional function proxy model to obtain and verify the final failure probability.

9. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the method for detecting the reliability of hull beam strength based on a learning function as claimed in claim 1.

10. A terminal device comprising a processor and a computer-readable storage medium, wherein the processor is configured to implement various instructions; and the computer-readable storage medium is configured to store a plurality of instructions, wherein: The instructions are suitable for being loaded by a processor and executed by a method for detecting the reliability of hull beam strength based on a learning function as claimed in claim 1.

Images