Method for analyzing structural reliability of liquid hydrogen storage tank of offshore platform
By constructing a high-fidelity finite element model, data augmentation, and optimized Kriging model, combined with Monte Carlo simulation, the problems of high computational cost and insufficient accuracy in the reliability analysis of liquid hydrogen storage tanks on offshore platforms were solved, achieving efficient and accurate reliability assessment.
Patent Information
- Application Number
- CN202511776764.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-03-03
AI Technical Summary
Existing technologies struggle to efficiently and accurately assess the reliability of liquid hydrogen storage tanks on offshore platforms, especially in extreme environments, where computational costs are high and models lack sufficient accuracy in predicting low-probability failure events.
A high-fidelity finite element model was constructed, data augmentation was performed using the SMOTE algorithm, the Kriging model was optimized using the Bayesian optimization algorithm, a failure-sensitive surrogate model was trained, and reliability analysis was performed using Monte Carlo simulation.
It enables efficient and accurate reliability assessment of liquid hydrogen storage tanks on offshore platforms under complex loads, improves the prediction accuracy and identification efficiency of low-probability failure events, and reduces computational costs.
Smart Images
Figure CN121598697A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of structural engineering and safety assessment technology, and in particular relates to a method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms. Background Technology
[0002] Liquid hydrogen, as a clean energy carrier, plays an important role in the future energy structure. Offshore platforms are key infrastructure for developing and utilizing marine resources and deploying renewable energy conversion (such as offshore wind power to produce hydrogen). Therefore, deploying large liquid hydrogen storage tanks on offshore platforms for hydrogen storage and transfer is of great strategic significance.
[0003] However, the service environment of offshore liquid hydrogen storage tanks is extremely harsh, and the technical challenges they face are fundamentally different from those of onshore tanks or other conventional hazardous materials storage tanks. The tank structure must simultaneously withstand extremely low temperatures (approximately 20K), internal high pressure, dynamic impacts from liquid hydrogen sloshing, and complex inertial loads transmitted by the six degrees of freedom motion of the offshore platform caused by wind, waves, and currents. Especially in critical areas such as welds, stress concentration, fatigue damage, and material property degradation are easily generated under the combined effects of liquid hydrogen phase change pressure fluctuations, temperature gradient changes, and external loads, potentially leading to multiple failure modes (such as leakage and fracture). These failure modes exhibit highly nonlinear coupling and dynamic evolution characteristics.
[0004] Currently, reliability assessment of liquid hydrogen storage tanks faces challenges, primarily due to the high computational cost of models. Accurate simulation of the tank's dynamic response under thermo-mechanical and fluid-structure interaction relies on complex finite element models (FEMs). However, probabilistic reliability analyses such as Monte Carlo simulations require a large number of samples, and the enormous time consumption of a single FEM calculation makes this approach impractically expensive and unfeasible in engineering. Secondly, constructing surrogate models presents difficulties. While surrogate models such as neural networks can be introduced to replace FEMs to address computational efficiency issues, constructing an effective surrogate model in this specific scenario requires solving two major challenges: first, the extreme sparsity of failure samples. Because tank failures are extremely low-probability events, the training data is severely imbalanced, resulting in standard models having almost no ability to learn and predict failure modes, posing a risk of missed detections; second, the highly nonlinear system response. The strong coupling of multiphysics fields creates complex relationships between input and output, making standard neural network models prone to training failures and inaccurate fitting of failure boundaries under the dual influence of this complexity and data sparsity. Finally, there are limitations in existing research methods. While some research has attempted to utilize backpropagation (BP) neural networks for analysis, they are typically introduced only as general models, without specifically addressing the aforementioned issues of data imbalance and highly nonlinear models. Existing solutions either ignore the impact of data imbalance or lack a systematic approach to optimize neural networks, enabling them to maintain high-accuracy prediction capabilities for extremely low-probability events under such demanding conditions. Summary of the Invention
[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a structural reliability analysis method for liquid hydrogen storage tanks on offshore platforms. This method solves the problems of reliability analysis model failure due to extremely sparse failure samples, low computational efficiency, and insufficient accuracy of reliability analysis models due to high nonlinearity. As a result, it enables efficient and accurate assessment of the structural reliability of liquid hydrogen storage tanks on offshore platforms under complex loads.
[0006] To achieve the above objectives, the technical solution adopted by this invention is: a method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms, comprising the following steps: S1. Construct a high-fidelity finite element model of the liquid hydrogen storage tank on the offshore platform; S2. Based on the high-fidelity finite element model, simulate the response of the liquid hydrogen storage tank under alternating loads to obtain the original dataset. S3. Based on the original dataset, perform data augmentation using the SMOTE algorithm for failure boundary augmentation to obtain the augmented dataset; S4. Based on the augmented dataset, using the Bayesian optimization algorithm and guided by a composite objective function that imposes high penalty weights on misclassification of failure samples, train and optimize the Kriging model to obtain the failure-sensitive surrogate model. S5. Based on the failure-sensitive proxy model and combined with Monte Carlo simulation, the reliability analysis results of the liquid hydrogen storage tank on the offshore platform are obtained.
[0007] To address the problems of high computational cost, insufficient prediction accuracy due to extremely sparse failure samples, and the lack of sensitivity of conventional models to predicting low-probability failure events in reliability assessment of liquid hydrogen storage tanks on offshore platforms under complex alternating loads, this invention proposes a structural reliability analysis method for liquid hydrogen storage tanks on offshore platforms, based on a Bayesian optimized Kriging model. First, an original dataset containing sparse failure samples is generated through finite element simulation. Then, the SMOTE algorithm, which enhances performance based on failure boundaries, is used for data balancing. Next, a Bayesian optimization process is guided by a failure prediction-weighted objective function to train a Kriging model highly sensitive to failure states. Finally, Monte Carlo simulation is combined to achieve efficient reliability assessment. This method improves the prediction accuracy and identification efficiency for low-probability liquid hydrogen storage tank failure events while reducing computational costs, providing reliable technical support for the safety design and risk control of liquid hydrogen storage tanks on offshore platforms.
[0008] Further: The specific steps of S1 include: S101. Obtain the mechanical and thermophysical properties of the materials used in liquid hydrogen storage tanks on offshore platforms at extremely low temperatures. S102. Based on the mechanical and thermophysical property parameters at extremely low temperatures, the constitutive relation of the material is obtained; S103. Based on the geometry, material constitutive relations, thermal insulation structure, and support structure of the liquid hydrogen storage tank on the offshore platform, a three-dimensional finite element model is established using finite element software. S104. Based on the three-dimensional finite element model, boundary conditions are added according to the installation method of the liquid hydrogen storage tank on the offshore platform to obtain a high-fidelity finite element model.
[0009] The further beneficial effects mentioned above are as follows: by obtaining the cryogenic material parameters and thermophysical performance parameters of the liquid hydrogen storage tank material on the offshore platform and establishing the material constitutive relationship, it is possible to consider the nonlinear, temperature-dependent mechanical behaviors of the material caused by the ductile-brittle transition, strength improvement, and fatigue curve changes due to the cryogenic environment; on this basis, a three-dimensional model is established by combining the actual geometric structure and support system of the storage tank, and by applying boundary conditions, a high-fidelity finite element model that can accurately reproduce the actual working conditions of the liquid hydrogen storage tank on the offshore platform is constructed, providing a real and reliable basis for reliability analysis and ensuring the credibility of the reliability assessment.
[0010] Further: The specific steps of S2 include: S201. Set alternating load conditions to cover extreme marine conditions for the high-fidelity finite element model to obtain a high-fidelity finite element model under alternating load. S202. Based on the high-fidelity finite element model under alternating load, stress distribution characteristics are analyzed through numerical simulation to obtain dynamic stress-strain response results under alternating load. S203. Based on the multi-mode failure criterion, the failure state is identified from the dynamic stress-strain response results under alternating load, and the failure modes under alternating load conditions are obtained. S204. Adjust the input parameters to include alternating load conditions covering extreme marine conditions, and repeat S201 to S203, recording the input parameters, dynamic stress-strain response results under alternating loads, and failure modes under alternating load conditions each time to obtain the original dataset.
[0011] The further beneficial effects mentioned above are as follows: By applying alternating loads including extreme offshore conditions, this invention can obtain the dynamic stress-strain response of liquid hydrogen storage tanks on offshore platforms under complex loads, and accurately identify the failure state based on multi-mode failure criteria; by repeatedly adjusting the input parameters to perform multi-condition simulations, a raw dataset containing catastrophic load inputs, stress-strain responses, and failure state mappings is finally constructed; the problem of insufficient failure samples leading to low quality of subsequent model training data in the prior art is solved, providing comprehensive basic data for subsequent data augmentation and model optimization, and improving the accuracy and reliability of reliability analysis.
[0012] Furthermore, the specific steps of S3 include: S301. Perform sample distribution analysis on the original dataset, count the proportion of failed samples to safe samples, and obtain the extreme sparsity ratio of failed samples. S302. Based on the extreme sparsity ratio of the failure samples, the SMOTE algorithm for failure boundary enhancement performs linear interpolation between the failure samples and their nearest neighbors of the same type in the feature space to obtain synthetic samples. S303. Add the synthetic samples to the original dataset until the number of failed samples and safe samples reaches a relative balance, thus obtaining the augmented dataset.
[0013] The further beneficial effects mentioned above are as follows: by analyzing the sample distribution, we can understand the extreme sparsity ratio of failure samples, which provides a basis for data augmentation. The SMOTE algorithm, which is guided by failure boundary augmentation, generates synthetic samples by interpolating between failure samples and their nearest neighbors of the same type. This can avoid the problem of synthetic samples being detached from physical reality in conventional methods. Finally, we obtain the augmented dataset through sample balancing, which solves the problem that the imbalance of the original data makes it difficult for the model to learn failure modes, and provides data support for the subsequent training of the Kriging model.
[0014] Furthermore, the specific steps of S4 include: S401. Select the Kriging model as the basic model and determine the composite objective function for weighting failure prediction based on the maritime scenario. S402. Based on the composite objective function weighted for failure prediction, the Bayesian optimization algorithm is used to perform a global search in the hyperparameter space to obtain the optimal hyperparameter combination of the Kriging model. S403. Configure the optimal hyperparameter combination of the Kriging model into the Kriging model to obtain the optimized Kriging model. S404. Based on the augmented dataset, train the optimized Kriging model to obtain the failure-sensitive agent model.
[0015] The further beneficial effects mentioned above are as follows: by selecting the Kriging model and combining it with the maritime scenario to determine the weighted composite objective function for failure prediction, the number of cases where failure samples are incorrectly predicted as safe is reduced; then, by using Bayesian optimization to perform a global search in the hyperparameter space, the optimal hyperparameter combination of the Kriging model is obtained; finally, the Kriging model is trained and optimized using an augmented dataset to obtain a failure-sensitive surrogate model, which solves the problem that conventional surrogate models are not sensitive to predicting low-probability failure events and reduces the risk of missed reporting of dangerous operating conditions.
[0016] Furthermore, the expression for the composite objective function is as follows:
[0017] in, For a composite objective function, For the prediction error of safe samples, The prediction error for failed samples, The weighting coefficients for the prediction error of safe samples. Let be the weighting coefficient of the prediction error for the failed samples, and .
[0018] The further beneficial effects mentioned above are: by using a composite objective function, the Bayesian optimization process prioritizes searching for hyperparameter combinations that can accurately identify sparse failure samples, enabling the failure-sensitive surrogate model to focus on low-probability failure events, thereby improving the predictive sensitivity and identification accuracy of the failure-sensitive surrogate model.
[0019] Furthermore, the specific steps of S5 include: S501. Determine the main random variables affecting the reliability of liquid hydrogen storage tanks, and formulate the probability distribution type and parameters of the main random variables according to the actual situation, so as to obtain the probability distribution of the main random variables. S502. Based on the probability distribution of the main random variables, perform Monte Carlo simulation sampling to obtain random variable samples. S503. Based on the random variable samples, use the failure-sensitive surrogate model to make predictions and output the failure status judgment result for each random variable sample. S504. Based on the failure status determination results of each random variable sample, count the number of random variable samples that are determined to be failed, and calculate the reliability analysis results including failure probability and reliability.
[0020] The further beneficial effects mentioned above are as follows: By combining the failure-sensitive surrogate model with Monte Carlo simulation, the high accuracy and high sensitivity of the failure-sensitive surrogate model can replace the traditional time-consuming finite element calculation, enabling rapid and accurate quantification of the failure probability of liquid hydrogen storage tanks under marine conditions. At the same time, it has high computational efficiency and high failure determination accuracy, improving the computational efficiency of reliability analysis and providing quantitative data support for the safety assessment of liquid hydrogen storage tanks on offshore platforms.
[0021] The beneficial effects of this invention are as follows: This invention solves the problems of low computational efficiency, model reliability analysis failure due to extremely sparse failure samples, and insufficient accuracy of reliability analysis due to high nonlinearity in traditional reliability analysis methods for liquid hydrogen storage tanks on offshore platforms. It achieves efficient and accurate assessment of the structural reliability of liquid hydrogen storage tanks on offshore platforms under complex loads. Based on the Bayesian optimization algorithm, this invention optimizes the Kriging model to obtain a failure-sensitive surrogate model, improving the prediction accuracy and efficiency of reliability assessment for key failure modes, and providing reliable technical support for the safety design and risk control of liquid hydrogen storage tanks on offshore platforms. The failure-sensitive surrogate model of this invention, combined with Monte Carlo simulation, can solve the problems of model failure due to extremely sparse failure samples and insufficient accuracy of reliability analysis due to high nonlinearity, significantly improving the efficiency and accuracy of structural reliability analysis of liquid hydrogen storage tanks on offshore platforms, and providing technical support for the safety design and operation and maintenance of liquid hydrogen storage tanks on offshore platforms. Attached Figure Description
[0022] Figure 1 This is a flowchart of a method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms. Detailed Implementation
[0023] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0024] Example 1 This invention provides a method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms, characterized by the following steps: S1. Construct a high-fidelity finite element model of the liquid hydrogen storage tank on the offshore platform; S2. Based on the high-fidelity finite element model, simulate the response of the liquid hydrogen storage tank under alternating loads to obtain the original dataset. S3. Based on the original dataset, perform data augmentation using the SMOTE algorithm for failure boundary augmentation to obtain the augmented dataset; S4. Based on the augmented dataset, using the Bayesian optimization algorithm and guided by a composite objective function that imposes high penalty weights on misclassification of failure samples, train and optimize the Kriging model to obtain the failure-sensitive surrogate model. S5. Based on the failure-sensitive proxy model and combined with Monte Carlo simulation, the reliability analysis results of the liquid hydrogen storage tank on the offshore platform are obtained.
[0025] In one embodiment of the present invention, step S1 specifically includes: S101. Obtain the mechanical and thermophysical property parameters of the materials used in the liquid hydrogen storage tanks of offshore platforms at extremely low temperatures. In a specific embodiment of the present invention, the extremely low temperature is selected to be around 20K. The mechanical property parameters include parameters such as elastic modulus, yield strength, tensile strength, and fatigue SN curve. The thermophysical property parameters include parameters such as thermal conductivity, specific heat capacity, and coefficient of thermal expansion. The mechanical and thermophysical property parameters of the present invention at extremely low temperatures take into account the nonlinear and temperature-dependent mechanical behaviors of the materials used in the liquid hydrogen storage tanks of offshore platforms, such as the ductile-brittle transition, strength improvement, and fatigue curve changes caused by the low temperature environment. Based on this, a material constitutive model that can describe the above properties is established, laying the foundation for subsequent numerical simulation. S102. Based on the mechanical and thermophysical property parameters at extremely low temperatures, the constitutive relation of the material is obtained; S103. Based on the geometric structure, material constitutive relationship, insulation structure, and support structure of the liquid hydrogen storage tank on the offshore platform, a three-dimensional finite element model is established using finite element software. The three-dimensional finite element model includes the physical structure of the inner liner, outer shell, insulation layer (such as vacuum interlayer, multi-layer insulation material, microsphere filling, etc.), support structure, and pipeline connection, and performs mesh generation, refining the mesh in stress concentration areas such as welds and openings. S104. Based on the three-dimensional finite element model, boundary conditions are added according to the installation method of the liquid hydrogen storage tank on the offshore platform to obtain a high-fidelity finite element model. The boundary conditions are determined by the installation method of the liquid hydrogen storage tank on the offshore platform, and displacement constraints corresponding to the installation method are added. In a specific embodiment of the present invention, taking the common horizontal liquid hydrogen storage tank double saddle installation method on the offshore platform as an example, the addition of boundary conditions follows the principle of one end fixed and one end sliding to adapt to the thermal shrinkage deformation of materials under the extremely low temperature environment of liquid hydrogen. For fixed-end support, select the bottom node of the saddle on one side of the storage tank and apply fully constrained boundary conditions. The constraint here simulates the rigid connection between the storage tank and the deck of the offshore platform, which is used to limit the rigid body displacement of the storage tank and serve as a reference point for thermal deformation.
[0026] For sliding end support, select the bottom node of the saddle on the other side of the storage tank and apply a vertical constraint. U y and horizontal U x Displacement constraints, but in the axial direction of the tank. U z The translational degrees of freedom are released (i.e., no constraints are applied). This boundary setting allows the tank to contract axially inward after being filled with liquid hydrogen and cooled, eliminating the huge thermal stress caused by the temperature difference.
[0027] Multi-point coupling constraints are used to ensure that the six degrees of freedom motion of the platform (acceleration, sway, and other inertial loads) applied in subsequent steps can be accurately transmitted to the tank body through these boundaries, considering that the deck of the offshore platform is not an absolutely rigid body. In the finite element model, the connection between the saddle base plate and the reference point of the platform deck is defined by multi-point constraints or contact pairs.
[0028] In one embodiment of the present invention, the specific steps of S2 include: S201. Set alternating load conditions to cover extreme marine conditions for the high-fidelity finite element model to obtain a high-fidelity finite element model under alternating load. This includes alternating load conditions covering extreme marine operating conditions, specifically including: Internal load characterizes the liquid hydrogen gravity and internal pressure of the liquid hydrogen storage tank on the offshore platform. The internal pressure takes into account pressure variations such as operating pressure, sloshing pressure, and evaporation pressure fluctuations. Temperature load characterizes the temperature gradient inside and outside the liquid hydrogen storage tank on the offshore platform, as well as the temperature changes during operation; External loads characterize the motion response of the offshore platform, such as acceleration and tilt angle, and the inertial force generated after being transferred to the liquid hydrogen storage tank, wind loads, wave loads, etc. By introducing random wave spectrum and wind spectrum, the six-degree-of-freedom motion time history of the offshore platform is simulated and generated. Alternating load is the most critical load, and the focus is on handling the randomness of load changes. The six-degree-of-freedom motion time history is applied as an inertial force to the high-fidelity finite element model of the liquid hydrogen storage tank. Alternating load is not a simple periodic load, but a complex dynamic excitation with randomness and wide bandwidth characteristics. It is coupled with the sloshing pressure and temperature stress inside the storage tank, and is a key factor leading to complex failure modes that are difficult to predict by conventional methods.
[0029] S202. Based on the high-fidelity finite element model under alternating load, stress distribution characteristics are analyzed through numerical simulation to obtain dynamic stress-strain response results under alternating load. In a specific embodiment of the present invention, static and transient thermo-mechanical coupling analysis is performed through stress distribution characteristic analysis to obtain the dynamic stress-strain response history of the liquid hydrogen storage tank under alternating load conditions, including extreme marine conditions, and the dynamic stress-strain response results of the liquid hydrogen storage tank under alternating load are obtained by organizing the data. S203. Based on the multi-mode failure criteria, the failure state is identified from the dynamic stress-strain response results under alternating loads to obtain the failure modes under alternating loads. Among them, the multi-mode failure criteria can use common techniques such as the Mises yield criterion, the maximum principal stress criterion, the fracture mechanics criterion, and the fatigue cumulative damage theory to determine the possible failure modes of the storage tank and to obtain the failure modes of the liquid hydrogen storage tank under alternating loads. S204. Adjust the input parameters to include alternating load conditions covering extreme marine conditions, and repeat S201 to S203, recording the input parameters, dynamic stress-strain response results under alternating loads, and failure modes under alternating load conditions each time to obtain the original dataset. Since errors are unavoidable in the manufacturing of liquid hydrogen storage tanks and the marine environment influences different liquid hydrogen storage tanks, different tanks have different performance parameters. Therefore, the input parameters to include alternating load conditions covering extreme marine conditions are adjusted, including parameters related to marine conditions, random parameters of tank material properties, and geometric defect sizes. Through multiple stress distribution characteristic analyses and failure mode analyses, several sets of different input parameters, dynamic stress-strain responses, and failure mode data are obtained, and integrated to obtain the original dataset.
[0030] In one embodiment of the present invention, step S3 specifically includes: S301. Perform sample distribution analysis on the original dataset, count the proportion of failed samples to safe samples, and obtain the extreme sparsity ratio of failed samples. S302. Based on the extreme sparsity ratio of failure samples, the SMOTE algorithm for failure boundary enhancement performs linear interpolation between the failure sample and its nearest neighbor of the same type in the feature space to obtain a synthetic sample; the synthetic sample can effectively describe and enhance the real failure boundary. S303. Add the synthetic samples to the original dataset until the number of failed samples and safe samples reaches a relative balance, thus obtaining the augmented dataset.
[0031] Existing technologies typically only introduce neural networks or conventional response surface models as general tools, ignoring the extremely sparse nature of liquid hydrogen storage tank failure samples, i.e., the characteristics of extremely low probability events. Conventional models use general fitting accuracy as the optimization objective. When the data is severely imbalanced, the model tends to obtain extremely high global accuracy by predicting all cases as safe, leading to serious underreporting of hazardous conditions. In contrast, this invention uses Bayesian optimization to specifically find hyperparameter combinations sensitive to failure modes. With the addition of a weighted objective function, this invention forces the Kriging model to shift its optimization focus from global fitting to failure boundary prediction, which is highly relevant to engineering safety. Combined with SMOTE enhancement technology, this improves the accuracy of identifying low-probability catastrophic failure events of liquid hydrogen storage tanks on offshore platforms under multi-field coupling, avoiding misjudgments of risks in the design.
[0032] In one embodiment of the present invention, step S4 specifically includes: S401. Selecting the Kriging model as the basic model, and determining the composite objective function for failure prediction weighting based on the marine scenario; conventional reliability analysis models for liquid hydrogen storage tanks on offshore platforms generally aim to reduce global errors, which ignores the prediction accuracy for a very small number of failure samples, resulting in insufficient accuracy and low reliability of the final analysis results. The failure prediction weighted composite objective function designed in this invention applies a much higher penalty weight to the case of incorrectly predicting failure samples as safe when evaluating the performance of the Kriging model, so as to improve the identification accuracy of a very small number of failure samples and improve the accuracy of reliability analysis. S402. Based on the composite objective function weighted for failure prediction, a Bayesian optimization algorithm is used to perform a global search in the hyperparameter space to obtain the optimal hyperparameter combination for the Kriging model. Specifically, the Bayesian optimization algorithm, guided by the composite objective function weighted for failure prediction, performs an efficient global search in the hyperparameter space of the Kriging model, including parameters such as covariance function type and correlation length, ultimately obtaining a set of optimal hyperparameter combinations that naturally give the Kriging model the highest predictive sensitivity to the failure region. The expression for the composite objective function is as follows:
[0033] in, For a composite objective function, For the prediction error of safe samples, The prediction error for failed samples, The weighting coefficients for the prediction error of safe samples. Let be the weighting coefficient of the prediction error for the failed samples, and This guides the model to prioritize optimizing the prediction accuracy of failed samples; S403. Configure the optimal hyperparameter combination of the Kriging model into the Kriging model to obtain the optimized Kriging model. S404. Based on the augmented dataset, the optimized Kriging model is trained to obtain a failure-sensitive surrogate model. The augmented dataset is divided into a training set and a test set. The optimized Kriging model is trained using the training set, and its generalization ability is evaluated using the test set. The final result is a surrogate model with fast prediction speed and extremely high accuracy in identifying combinations of input parameters that may lead to failure; this is the failure-sensitive surrogate model. Specifically, S403 utilizes the mathematical principles of Kriging regression in existing technologies. The optimal hyperparameter combination obtained in S402 is substituted into the Kriging model structure as a known constant. The augmented dataset is used as training sample points. By calculating the correlation matrix and its inverse matrix between the training sample points, the weighting coefficients of the Kriging model are determined, thereby establishing a highly nonlinear mathematical mapping relationship between input parameters such as complex alternating loads and material parameters and the output response. This completes the construction of the failure-sensitive surrogate model. The resulting failure-sensitive surrogate model is essentially an optimized mathematical function capable of interpolating and predicting any new input point in the feature space, exhibiting particularly high sensitivity to low-probability failure regions.
[0034] In one embodiment of the present invention, step S5 specifically includes: S501. Determine the main random variables affecting the reliability of liquid hydrogen storage tanks, and formulate the probability distribution type and parameters of the main random variables according to the actual situation to obtain the probability distribution of the main random variables; among them, the main random variables include the boundary conditions of the marine environment and various loads. S502. Based on the probability distribution of the main random variables, perform Monte Carlo simulation sampling to obtain random variable samples. S503. Based on the random variable samples, use the failure-sensitive surrogate model to make predictions and output the failure status judgment result for each random variable sample. S504. Based on the failure status determination results of each random variable sample, count the number of random variable samples that are determined to be failed, and calculate the reliability analysis results including failure probability and reliability.
[0035] The beneficial effects of this invention are as follows: This invention can solve the problems of low computational efficiency, model reliability analysis failure due to extremely sparse failure samples, and insufficient accuracy of reliability analysis due to high nonlinearity in traditional reliability analysis methods for offshore platform liquid hydrogen storage tanks; it achieves efficient and accurate assessment of the structural reliability of offshore platform liquid hydrogen storage tanks under complex loads. This invention optimizes the Kriging model based on the Bayesian optimization algorithm to obtain a failure-sensitive surrogate model, improving the prediction accuracy and efficiency of reliability assessment for key failure modes, and providing reliable technical support for the safety design and risk control of offshore platform liquid hydrogen storage tanks; the failure-sensitive surrogate model of this invention, combined with Monte Carlo simulation, can solve the problems of model failure due to extremely sparse failure samples and insufficient accuracy of reliability analysis due to high nonlinearity, significantly improving the efficiency and accuracy of structural reliability analysis of offshore platform liquid hydrogen storage tanks, and providing technical support for the safety design and operation and maintenance of offshore platform liquid hydrogen storage tanks.
Claims
1. A method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms, characterized in that, Includes the following steps: S1. Construct a high-fidelity finite element model of the liquid hydrogen storage tank on the offshore platform; S2. Based on the high-fidelity finite element model, simulate the response of the liquid hydrogen storage tank under alternating loads to obtain the original dataset. S3. Based on the original dataset, perform data augmentation using the SMOTE algorithm for failure boundary augmentation to obtain the augmented dataset; S4. Based on the augmented dataset, using the Bayesian optimization algorithm and guided by a composite objective function that imposes high penalty weights on misclassification of failure samples, train and optimize the Kriging model to obtain the failure-sensitive surrogate model. S5. Based on the failure-sensitive proxy model and combined with Monte Carlo simulation, the reliability analysis results of the liquid hydrogen storage tank on the offshore platform are obtained.
2. The method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms according to claim 1, characterized in that, The specific steps of S1 include: S101. Obtain the mechanical and thermophysical properties of the materials used in liquid hydrogen storage tanks on offshore platforms at extremely low temperatures. S102. Based on the mechanical and thermophysical property parameters at extremely low temperatures, the constitutive relation of the material is obtained; S103. Based on the geometry, material constitutive relations, thermal insulation structure, and support structure of the liquid hydrogen storage tank on the offshore platform, a three-dimensional finite element model is established using finite element software. S104. Based on the three-dimensional finite element model, boundary conditions are added according to the installation method of the liquid hydrogen storage tank on the offshore platform to obtain a high-fidelity finite element model.
3. The method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms according to claim 1, characterized in that, The specific steps of S2 include: S201. Set alternating load conditions to cover extreme marine conditions for the high-fidelity finite element model to obtain a high-fidelity finite element model under alternating load. S202. Based on the high-fidelity finite element model under alternating load, stress distribution characteristics are analyzed through numerical simulation to obtain dynamic stress-strain response results under alternating load. S203. Based on the multi-mode failure criterion, the failure state is identified from the dynamic stress-strain response results under alternating load, and the failure modes under alternating load conditions are obtained. S204. Adjust the input parameters to include alternating load conditions covering extreme marine conditions, and repeat S201 to S203, recording the input parameters, dynamic stress-strain response results under alternating loads, and failure modes under alternating load conditions each time to obtain the original dataset.
4. The method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms according to claim 1, characterized in that, The specific steps of S3 include: S301. Perform sample distribution analysis on the original dataset, count the proportion of failed samples to safe samples, and obtain the extreme sparsity ratio of failed samples. S302. Based on the extreme sparsity ratio of the failure samples, the SMOTE algorithm for failure boundary enhancement performs linear interpolation between the failure samples and their nearest neighbors of the same type in the feature space to obtain synthetic samples. S303. Add the synthetic samples to the original dataset until the number of failed samples and safe samples reaches a relative balance, thus obtaining the augmented dataset.
5. The method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms according to claim 1, characterized in that, The specific steps of S4 include: S401. Select the Kriging model as the basic model and determine the composite objective function for weighting failure prediction based on the maritime scenario. S402. Based on the composite objective function weighted for failure prediction, a Bayesian optimization algorithm is used to perform a global search in the hyperparameter space to obtain the optimal hyperparameter combination of the Kriging model. S403. Configure the optimal hyperparameter combination of the Kriging model into the Kriging model to obtain the optimized Kriging model. S404. Based on the augmented dataset, train the optimized Kriging model to obtain the failure-sensitive agent model.
6. The method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms according to claim 5, characterized in that, The expression for the composite objective function is as follows: in, For a composite objective function, For the prediction error of safe samples, The prediction error for failed samples, The weighting coefficients for the prediction error of safe samples. Let be the weighting coefficient of the prediction error for the failed samples, and .
7. The method for structural reliability analysis of liquid hydrogen storage tanks on offshore platforms according to claim 1, characterized in that, The specific steps of S5 include: S501. Determine the main random variables affecting the reliability of liquid hydrogen storage tanks, and formulate the probability distribution type and parameters of the main random variables according to the actual situation, so as to obtain the probability distribution of the main random variables. S502. Based on the probability distribution of the main random variables, perform Monte Carlo simulation sampling to obtain random variable samples. S503. Based on the random variable samples, use the failure-sensitive surrogate model to make predictions and output the failure status judgment result for each random variable sample. S504. Based on the failure status determination results of each random variable sample, count the number of random variable samples that are determined to be failed, and calculate the reliability analysis results including failure probability and reliability.