A welding structure fatigue life reliability evaluation method based on a physical information neural network

Abstract

The application discloses a welding structure fatigue life reliability evaluation method based on a physical information neural network, which is applied to the technical field of structure life prediction, and aims at accurately evaluating the fatigue life and reliability of a welding structure under the coupling influence of manufacturing process and service load double uncertainties in actual engineering, which is still a difficult problem to be solved urgently. The application characterizes and quantifies multiple-source uncertainties by systematically analyzing a geometric structure and a service load, constructs a finite element model considering the uncertainties and performs dynamic crack propagation simulation to obtain key mechanical response parameters. Furthermore, the application fuses numerical simulation and test data to establish a PINN fatigue life prediction model with the Paris formula as a physical constraint, and finally realizes reliability evaluation based on subset simulation.

Description

Technical Field

[0001] This invention belongs to the field of structural life prediction, and specifically relates to a reliability assessment technology for fatigue life of high-strength steel boom structures that considers the coupling effect of welding process and service load. Background Technology

[0002] Welded structures, especially high-strength steel boom structures, are widely used in key fields such as bridges, engineering machinery, shipbuilding and marine engineering, and heavy equipment due to their excellent mechanical properties. Under long-term, complex alternating service loads, fatigue failure is one of the main failure modes for these structures, directly affecting the overall structural safety and durability. Currently, the assessment and prediction of fatigue life of welded structures in engineering mainly relies on analytical or numerical methods based on fracture mechanics, which typically employ or modify classical fatigue crack propagation models (such as the Paris formula). These methods generally have a significant limitation: they fail to systematically consider the complex coupling effects and uncertainties between manufacturing processes and service loads.

[0003] Specifically, the welding process directly determines the microstructure, residual stress distribution, geometric discontinuities (such as weld toe shape and undercut), and initial defects in the joint area. These initial states introduced by the manufacturing process exhibit significant statistical dispersion. Once the structure enters service, its fatigue damage evolution is the result of the interaction between external cyclic loads and the aforementioned process characteristics. For example, high tensile residual stress significantly lowers the fatigue threshold and accelerates crack initiation; specific weld toe geometry exacerbates local stress concentration, which, combined with service loads in specific directions, greatly shortens crack propagation life. This nonlinear interaction mechanism between process characteristics and service loads is key to determining the actual fatigue performance of welded structures.

[0004] However, traditional methods often treat these two aspects separately: either simplifying the process influence to a fixed safety factor, or only performing probabilistic statistics on the load while ignoring the uncertainty of its association with process parameters. Existing physical model-based methods struggle to construct analytical equations that accurately describe this coupling mechanism; while purely data-driven methods, although capable of discovering correlations from large amounts of data, are highly dependent on complete coupled datasets and lack physical constraints, resulting in poor extrapolation of prediction results.

[0005] Therefore, accurately assessing the fatigue life and reliability of welded structures under the coupled influence of uncertainties in manufacturing processes and service loads remains a pressing problem in practical engineering. This limits the accuracy of life prediction and affects the scientific and economical nature of maintenance strategies. In recent years, Physical Information Neural Networks (PINNs), as an emerging fusion modeling technology, have shown potential in handling such complex problems. They can embed the physical laws describing the fatigue process into the learning process as constraints, and are expected to learn and characterize the coupled influence of process parameters and load history on damage evolution with limited data. Constructing a method that can systematically characterize the entire coupling mechanism from manufacturing uncertainties to the randomness of service loads, and effectively integrate PINN technology to achieve efficient reliability quantification, has significant theoretical and practical value for improving the accuracy and engineering practicality of fatigue life assessment of welded structures. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention proposes a fatigue life reliability assessment method for welded structures based on physical information neural networks. By constructing a life prediction model that considers the uncertainties of welding process, material parameters, and geometric dimensions, the method achieves fatigue life prediction under multi-source uncertainties, resulting in low complexity and high prediction accuracy.

[0007] The technical solution adopted in this invention is: a method for assessing the fatigue life reliability of welded structures based on physical information neural networks, comprising:

[0008] S1. Analyze the geometry, load characteristics and main failure modes of the welded structure during service.

[0009] S2. Based on the analysis results of step S1, identify and quantify the key uncertainty factors affecting the fatigue reliability of the boom structure;

[0010] S3. Based on the key uncertainty factors identified and quantified in step S2, construct a finite element model of the welded structure and conduct dynamic crack propagation simulation analysis. Integrate the experimental results and finite element numerical simulation results to obtain the experimental results and finite element numerical simulation results.

[0011] S4. Establish a fully connected PINN neural network model, using the Paris formula for crack propagation in fracture mechanics as the model constraint. Calculate the fatigue crack propagation life based on experimental results and finite element simulation results. Use the experimental results and finite element simulation results as the input to the PINN neural network model, and the fatigue crack propagation life as the output of the PINN neural network model. Establish the relationship between crack damage and fatigue cycle in welded structures. Use automatic differential calculation to solve the gradient of the PINN neural network model and adjust the weights of the PINN neural network model.

[0012] S5. Based on the PINN neural network model trained in step S4, the predicted fatigue life of the welded structure to be analyzed is obtained, and the fatigue reliability assessment model of the welded structure to be analyzed is obtained based on the predicted fatigue life.

[0013] The beneficial effects of this invention are as follows: The evaluation method provided by this invention has the following beneficial effects: By systematically analyzing the geometric structure and service loads, multi-source uncertainties are characterized and quantified, a finite element model considering uncertainties is constructed and dynamic crack propagation simulation is performed to obtain key mechanical response parameters; then, numerical simulation and experimental data are integrated to establish a PINN fatigue life prediction model with the Paris formula as the physical constraint, and finally, reliability assessment is achieved based on subset simulation. This method can comprehensively consider the coupled influence of multiple factors such as welding process, material properties, service loads and geometric dimensions, and while ensuring the physical meaning of the model, it significantly improves the prediction accuracy and engineering generalization ability, providing a complete and practical solution for the fatigue life reliability assessment of welded structures. Attached Figure Description

[0014] Figure 1 This is a flowchart of a method for assessing the fatigue life reliability of welded structures based on a Physical Information Neural Network (PINN) according to the present invention.

[0015] Figure 2 This is a finite element simulation of the boom in an embodiment of the present invention.

[0016] Figure 3 This is a schematic diagram of the arrangement of the sliders between the tower arms in an embodiment of the present invention.

[0017] Figure 4 This is a schematic diagram of the arrangement of the sliders between the main arms in an embodiment of the present invention.

[0018] Figure 5 This is a schematic diagram of the amplitude-changing mechanism in an embodiment of the present invention.

[0019] Figure 6 This is a schematic diagram of the load on the boom in an embodiment of the present invention.

[0020] Figure 7 This is a schematic diagram of the fatigue crack propagation path in an embodiment of the present invention.

[0021] Figure 8 This is a schematic diagram of the architecture of the PINN model of the present invention.

[0022] Figure 9 This is a schematic diagram of the training results of the PINN model of the present invention. Detailed Implementation

[0023] To facilitate understanding of the technical content of this invention by those skilled in the art, the following description, in conjunction with the accompanying drawings, further illustrates the invention.

[0024] like Figure 1 The flowchart of the fatigue reliability assessment method for high-strength steel boom structures under the coupling effect of welding process and service load of the present invention is shown below. The specific steps are as follows:

[0025] S1. Analyze the geometry, load characteristics, and main failure modes of the welded structure during service.

[0026] The boom structure geometry analysis includes the analysis of boom composition, the structure of each boom section, and the connection method of each boom section. The boundary conditions for subsequent finite element simulation are determined through the boom structure analysis.

[0027] The service load analysis includes boom operation posture analysis and service load analysis under various working conditions.

[0028] The primary failure mode is fatigue fracture when crack propagation reaches a critical value.

[0029] S2. Systematically identify and quantify the multi-source uncertainties affecting the fatigue life of welded structures.

[0030] Identify and quantify the key sources of uncertainty affecting the fatigue reliability of the boom structure. The multi-source uncertainties mainly include: (1) process uncertainty: mainly refers to the residual stress field introduced into the structure by the welding manufacturing process (such as welding parameters, heat input, post-weld treatment, etc.) and the initial crack length that may lead to the initiation and propagation of fatigue cracks. (2) load uncertainty: refers to the randomness of the external load spectrum that the boom is subjected to during service, including the uncertainty of load amplitude, frequency and location, and the size of the plastic zone after the load is applied. (3) geometric uncertainty: refers to the dispersion of the boom's geometric dimensions caused by manufacturing tolerances. (4) material property uncertainty: refers to the uncertainty caused by subtle differences in material properties such as material strength and hardness.

[0031] The cross-sectional dimensions, initial crack length, and material properties determined in this step are used for the construction of the subsequent finite element model, and the load uncertainties determined in this step are used for the application of loads in the subsequent crack propagation simulation.

[0032] S3. Construct a finite element model of the welded structure based on the quantification results of multi-source uncertainties, and conduct dynamic crack propagation simulation analysis.

[0033] Based on the key uncertainty parameters determined in step S2, a mechanism affecting the fatigue life of high-strength steel boom structures is constructed. 3D random vector The experiment design employed Latin square sampling. Under the same temperature, a finite element model of the welded structure was established by inputting material parameters of different initial crack lengths, cross-sectional dimensions, weld seams, and base materials. Fatigue crack propagation numerical simulations were then performed based on the finite element model to obtain the mechanical response parameters of the material in each test. By integrating experimental and numerical simulation results, the material parameters for crack propagation under different finite element models were obtained. and Substituting the range of stress intensity factor in the mechanical response parameters into the Paris formula for integration The fatigue crack propagation life was obtained. The elastic modulus of the material, The yield strength of the material. Material parameters , Material parameters , The initial crack length is... For the thickness of the boom, This refers to the length of the boom.

[0034] The mechanical response parameters include: the range of stress intensity factor at the crack tip. , Type stress intensity factor , Type stress intensity factor , Type stress intensity factor , Integral length and crack propagation length; the values ​​of these mechanical response parameters can all be obtained through dynamic crack propagation simulation.

[0035] S4. Establish a fatigue life prediction model based on PINN by integrating numerical simulation and experimental data.

[0036] Building such Figure 8 As shown, the PINN deep learning model incorporates the physical equation of crack propagation rate, constructed in step S3. 3D random vector Using fatigue life as input and a composite loss function as output, the PINN model parameters are trained to predict the fatigue life of high-strength steel welded booms.

[0037] S5. Fatigue reliability assessment of high-strength steel boom structures based on subset simulation method.

[0038] Determine the factors affecting the fatigue life of high-strength steel boom structures 3D random vector as well as The relationship between 3D random vectors and the design life of high-strength steel welded structures is established, and a fatigue reliability analysis model for high-strength steel boom structures is created, yielding the function of failure for high-strength steel boom structures. The fatigue reliability of high-strength steel boom structures was estimated. The elastic modulus of the material, The yield strength of the material. Material parameters , Material parameters , The initial length of the crack. For the thickness of the boom, This refers to the length of the boom.

[0039] In this embodiment, step S1 is specifically as follows:

[0040] The boom selected in this embodiment includes a six-section tower boom and a three-section main boom. The first tower boom section has a length of 14470mm, the second tower boom section has a length of 14450mm, the third tower boom section has a length of 14430mm, the fourth tower boom section has a length of 14395mm, the fifth tower boom section has a length of 14405mm, and the sixth tower boom section has a length of 14710mm. The finite element model of the boom is as follows: Figure 2 As shown. Both the tower jib and the main boom adopt a telescopic structure. The tower jibs are connected to each other, and the main booms are connected to each other, using sliding blocks. The arrangement of the sliding blocks between the tower jibs is shown in [details omitted]. Figure 3 As shown, the arrangement of the sliders between the main booms is as follows: Figure 4 As shown. The tower jib and main jib are connected by a luffing mechanism, which is shown in the diagram. Figure 5 As shown.

[0041] The boom is subjected to rated load, wind load, and its own weight. Rated load includes the weight of the platform and boom, platform load, manual operating force, and wind load on the boom, platform, and platform load. Taking the maximum extension angle of the tower boom and the maximum extension angle of the main boom as an example, the load diagram is shown below. Figure 6 As shown, the boom is connected to the end of the last main boom section.

[0042] In this embodiment, step S2 is specifically as follows:

[0043] The multi-source uncertainties include: the influence of welding process on welded structure, the influence under different service loads, the influence of different geometries, and the influence of different material properties.

[0044] The impact of welding processes on welded structures includes: residual stress fields introduced into the structure by welding parameters, heat input, and post-weld treatment, and the initial crack length that may lead to fatigue crack initiation and propagation; different geometries include: the dispersion of geometric dimensions caused by manufacturing tolerances; different service loads refer to the randomness of the external load spectrum experienced by the boom during service, including load amplitude, frequency, and location of application; different material properties refer to the uncertainties caused by subtle differences in material properties such as strength and hardness, and this invention simulates these uncertainties through sampling.

[0045] The material parameter distribution is shown in Table 1:

[0046] Table 1 Random distribution of parameters

[0047]

[0048] Based on the material property values ​​in Table 1, several values ​​for each material property were obtained through sampling and uncertainty simulation. Since the initial crack length in engineering is generally 0.5-2 mm, this invention selects initial crack lengths of 0.5 mm, 1 mm, 1.5 mm, and 2 mm for simulation analysis. The quantified data is then used as training data for the subsequent PINN neural network model.

[0049] The load spectrum compiled by this invention based on actual measured data is as follows:

[0050] Table 2 Load spectrum of boom system

[0051]

[0052] In this embodiment, step S3 is specifically as follows:

[0053] S31, Finite element simulation of stress-displacement field of high-strength steel welded structure;

[0054] The finite element model of the high-strength steel welded structure was meshed using hexahedral elements, and static finite element simulation was performed to obtain the nodal displacement and stress distribution results; the mesh size of the finite element model was 50 mm.

[0055] S32, Adopt The stress intensity factor is calculated using the integral method.

[0056] Integrals and The relationship between the integrals is as follows:

[0057]

[0058]

[0059]

[0060]

[0061] in, express The integration path of the integral. Represents the true stress-displacement field of integral, The stress is the stress in the true stress-displacement field. For strain in the true stress-displacement field, The displacement is the actual stress-displacement field. Represents the additional stress-displacement field of integral, The stress is the stress in the additional stress-displacement field. For strain in the additional stress-displacement field, The displacement is the displacement of the additional stress-displacement field. For the integration path The cosine of the direction of the outward normal. Let Kronecker function be used when hour, ,when hour, . The strain energy density represents the actual stress-displacement field. This represents the strain energy density of the additional stress-displacement field. This represents the mutual integration of the real stress-displacement field and the additional stress-displacement field. The expression is as follows:

[0062]

[0063] Among them, superscript Represents the true stress-displacement field, superscript Represents the additional stress-displacement field. Represents the stress tensor. Represents the strain tensor. Represents the displacement vector. This refers to the Kronecker function. Describes an auxiliary function. Represents the first in Cartesian coordinates Each coordinate axis Corresponding to the x, y, and z axes respectively. This represents the first coordinate axis, which in fracture mechanics is taken as the direction of crack propagation. Represents the volume of the integration region. ;

[0064] The relationship between the integral and the stress intensity factor is as follows:

[0065]

[0066] in, Represents Poisson's ratio. Indicates the elastic modulus. The stress intensity factor represents the stress intensity factor of an open crack. The stress intensity factor represents the stress intensity factor of a slip crack. The stress intensity factor represents the tearing crack.

[0067] The expressions for the three types of stress intensity factors are as follows:

[0068]

[0069]

[0070]

[0071] S33. The crack propagation direction adopts the maximum circumferential stress criterion.

[0072] The maximum circumferential stress criterion specifies the relationship between crack propagation direction and maximum circumferential stress. The directions are the same. The expression is as follows:

[0073]

[0074] Crack propagation angle (direction) Depend on and The expression for the crack initiation angle is as follows:

[0075]

[0076] S34. Crack propagation increment adopts the POWER LAW rule;

[0077] The crack propagation increment, i.e., the crack propagation length at each propagation step, is expressed as follows:

[0078]

[0079] in, Indicates the amount of crack propagation. This indicates the median crack propagation. Represents a node The stress intensity factor amplitude at that location, This represents the stress intensity factor amplitude at the median point. and These are the increment values ​​set for each expansion step. Generally, the smaller the increment value, the better the result. curve, The smaller the error of the curve, the more the incremental value of each expansion step can be set according to actual needs in practical applications. When a certain expansion step corresponds to... When one of the three types calculated according to formulas (7)-(9) reaches the fracture toughness, the propagation stops; at this time, when the fracture toughness is reached, in The value 'a' corresponding to the curve is the final crack length.

[0080] Finite element simulation of fatigue crack propagation in S35 high-strength steel welded structure;

[0081] An initial crack is implanted into the finite element model, which is then divided into a sub-model containing the initial crack and other parts of the model. The mesh at the crack tip is refined. Nodes at the contact surfaces between the sub-model and other parts of the model are retained, and the nodal displacements at these surfaces are used as boundary conditions for the sub-model. Fatigue crack propagation finite element simulation is then performed to obtain the crack length and stress intensity factor. Curve, crack length, number of cycles The curve and the stress field distribution near the crack tip in each propagation step. The crack propagation path can be determined from the above theory; the propagation path of fatigue cracks is shown in [reference needed]. Figure 7 As shown.

[0082] In this embodiment, step S4 is specifically as follows:

[0083] A fully connected PINN neural network model is established, consisting of one input layer, four fully connected layers, and one output layer. The model inputs are material properties, initial crack length, and parameters from the Paris formula. and The boom cross-sectional dimensions are given, and the output is fatigue life. The Paris formula for crack propagation in fracture mechanics is used as the model constraint. Physical information is embedded into the network through a composite loss function, and experimental results (basic material parameters) are incorporated. and Using experimental data and finite element simulation results (material properties, initial crack length, cross-sectional dimensions) as inputs, and fatigue life as output, the relationship between crack damage and fatigue cycles in welded structures is established. The network gradient is solved automatically using differential calculus, and the network weights are adjusted. The architecture of this PINN model is shown below. Figure 8 ;

[0084] The number of iterations for each expansion step is obtained based on the Paris formula, which is expressed as follows:

[0085]

[0086] in, Indicates the crack length; This indicates the number of fatigue cycles, i.e., the fatigue crack propagation life. and Indicates the basic parameters of the material;

[0087] Integrating the Paris formula yields the fatigue crack propagation life formula as follows:

[0088]

[0089] in, Indicates the initial length of the prefabricated crack. Final crack length, Indicates the range of stress intensity factor. and Crack length and stress intensity factor were both simulated using finite element method. The curve is obtained;

[0090] The composite loss function is specifically expressed as follows:

[0091]

[0092] in, and These represent the weighting coefficients for the data fitting loss and the physical information loss. Let PINN be the loss function. This represents the data fitting loss component. The physical information loss component is expressed as follows:

[0093]

[0094]

[0095] Where MSE represents the mean squared error, and MSELoss represents the mean squared error between the two quantities. This indicates the number of periods predicted by the model. This represents the number of cycles in the finite element simulation. This represents the crack propagation rate calculated using PINN automatic differentiation. This represents the crack propagation rate calculated based on the Paris formula.

[0096] Material properties, initial crack length, and Paris formula parameters and As input to the PINN model, the Paris formula serves as the physical information error term, and fatigue life is the output. The training results of the PINN model of this invention are as follows: Figure 9 As shown. For sample 1, the fatigue life value obtained from numerical simulation is 113841, and the predicted value of the PINN model in this invention is 112567. For sample 2, the fatigue life value obtained from numerical simulation is 89451, and the predicted value of the PINN model in this invention is 88543.

[0097] In this embodiment, step S5 is specifically as follows:

[0098] Based on the PINN fatigue life prediction model established in step 4, a fatigue reliability assessment model for welded structures is constructed. The limit state function is defined. ,in The input variable vector includes uncertain parameters such as material properties, geometric dimensions, and load conditions. For a specified service life, the failure probability is defined as follows: When PINN predicts fatigue life Less than the target lifespan Time indicates structural failure. The failure domain is... Decomposed into a sequence of nested subsets: ,in Threshold sequence The corresponding failure probability decomposes as follows: By setting an appropriate intermediate threshold This transforms the problem of calculating the minimum failure probability into calculating the product of a series of larger conditional probabilities.

[0099] from prior distribution generate For each sample, calculate the limit state function. In this embodiment, Defined as the difference between the predicted fatigue life and the specified service life, i.e. .estimate ,in, To meet Number of samples. Threshold Selected as the corresponding samples of Quantiles for each level From the conditional distribution generate A sample, with Using existing samples as seeds, new samples are generated. These new samples are then input into the PINN neural network model to obtain new predicted fatigue life. The fatigue life is then calculated based on these new predicted fatigue life. ,estimate ,in, To meet Number of samples. Threshold Select as sample of Quantiles. When the threshold Stop at this time The failure probability is estimated as follows:

[0100]

[0101] , The value range is 1000-10000, and the value in this embodiment is 5000.

[0102] Those skilled in the art will recognize that the embodiments described herein are for the purpose of helping to understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A method for assessing the fatigue life reliability of welded structures based on physical information neural networks, characterized in that, include: S1. Analyze the geometry, load characteristics and main failure modes of the welded structure during service. S2. Based on the analysis results of step S1, identify and quantify the key uncertainty factors affecting the fatigue reliability of the boom structure; S3. Based on the key uncertainty factors identified and quantified in step S2, construct a finite element model of the welded structure and conduct dynamic crack propagation simulation analysis. Integrate the experimental results and finite element numerical simulation results to obtain the experimental results and finite element numerical simulation results. S4. Establish a fully connected PINN neural network model, using the Paris formula for crack propagation in fracture mechanics as the model constraint. Calculate the fatigue crack propagation life based on experimental results and finite element simulation results. Use the experimental results and finite element simulation results as the input to the PINN neural network model, and the fatigue crack propagation life as the output of the PINN neural network model. Establish the relationship between crack damage and fatigue cycle in welded structures. Use automatic differential calculation to solve the gradient of the PINN neural network model and adjust the weights of the PINN neural network model. S5. Based on the PINN neural network model trained in step S4, the predicted fatigue life of the welded structure to be analyzed is obtained, and the fatigue reliability assessment model of the welded structure to be analyzed is obtained based on the predicted fatigue life.

2. The fatigue life reliability assessment method for welded structures based on physical information neural networks according to claim 1, characterized in that, Key uncertainties in step S2 include: initial crack length, material properties, cross-sectional dimensions, and random load spectrum.

3. The fatigue life reliability assessment method for welded structures based on physical information neural networks according to claim 2, characterized in that, Step S3 specifically includes: Factors influencing the fatigue life of high-strength steel boom structures include initial crack length, material properties, and cross-sectional dimensions. 3D random vector distributed; This represents the total number of parameters corresponding to the initial crack length and material properties, where T denotes transpose; Experimental design was carried out using Latin square sampling. Under the same temperature, a finite element model of the welded structure was established by inputting material property parameters of different initial crack lengths, different cross-sectional dimensions, different welds, and base materials. Based on the finite element model, combined with the application of load by random load spectrum, fatigue crack propagation numerical simulation is carried out to obtain the mechanical response parameters of the material in each test. Material property parameters for crack propagation under different finite element models were obtained through experiments. and .

4. The fatigue life reliability assessment method for welded structures based on physical information neural networks according to claim 3, characterized in that, The fatigue life calculation process in step S4 is as follows: The range of stress intensity factor in the mechanical response parameters and the material parameters obtained from the experiment. and Substituting into the Paris formula and integrating, we obtain the fatigue crack propagation life: ； in, Indicates the initial length of the prefabricated crack. Final crack length, Indicates the range of stress intensity factor.

5. The fatigue life reliability assessment method for welded structures based on physical information neural networks according to claim 4, characterized in that, In step S4, the PINN neural network model includes: one input layer, four fully connected layers, and one output layer.

6. The fatigue life reliability assessment method for welded structures based on physical information neural networks according to claim 5, characterized in that, The PINN neural network model in step S4 uses a composite loss function during training, expressed as: ； in, and These represent the weighting coefficients for the data fitting loss and the physical information loss. Let PINN be the loss function. This represents the data fitting loss component. This represents the portion of physical information lost. , The expression is as follows: ； ； Where MSE represents the mean squared error, and MSELoss represents the mean squared error between the two quantities. This represents the fatigue crack propagation life predicted by the model. This represents the fatigue crack propagation life in finite element simulation. This represents the crack propagation rate calculated using PINN automatic differentiation. This represents the crack propagation rate calculated based on the Paris formula.

7. The fatigue life reliability assessment method for welded structures based on physical information neural networks according to claim 6, characterized in that, The implementation process of obtaining the fatigue reliability assessment model of the welded structure to be analyzed based on predicted fatigue life in step S5 is as follows: Define the limit state function , To specify the service life; Define the failure probability as ; When analyzing the predicted fatigue life of the welded structure Less than This indicates structural failure; failure domain Decomposed into a sequence of nested subsets: ,in Threshold sequence ; The corresponding failure probability decomposition is as follows: ; From the prior distribution generate For each sample, calculate the limit state function. , Defined as ; estimate ,in, To meet Number of samples, threshold Select as sample of quantiles; For the A subset of failure domains From the conditional distribution generate A sample, with Using existing samples as seeds, new samples are generated, and calculations are performed. ; estimate ,in, To meet The number of samples; threshold Select as sample of Quantiles, when the threshold Stop at this time The failure probability is estimated as follows: 。

Images