Stress Intensity Factor Transfer Function Prediction Method and System

Abstract

The present invention discloses a method and system for predicting a stress intensity factor transfer function, relating to the field of artificial intelligence technology. The method includes: after processing data on stress intensity factors, crack information, and wave frequencies, training a multi-layer perceptron model to obtain a multi-layer perceptron model capable of predicting stress intensity factors under different regular waves; constructing a stress intensity factor transfer function based on the predicted stress intensity factors to calculate the crack growth driving force, and calculating the corresponding crack growth amount based on the crack growth driving force; statistically obtaining the crack growth life curve under random sea conditions by counting the short-term sea condition experience time; statistically analyzing the fatigue life of an ocean structure based on the stress intensity factor transfer function under different crack sizes, and comparing the statistical results with the finite element results to obtain the corresponding fatigue life of the ocean structure, which helps to solve the problem that the prior art cannot efficiently and accurately statistically analyze the fatigue life of ocean structures.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and particularly to a method and system for predicting a stress intensity factor transfer function. Background Art

[0002] In the field of ocean engineering, ocean structures such as offshore platforms and aquaculture cages are subjected to complex ocean environments for a long time. Fatigue failure is a key factor affecting their safety and service life. Accurately counting the fatigue life of ocean structures is crucial for ensuring their normal operation and preventing accidents.

[0003] Existing fatigue life statistical methods have many limitations. On the one hand, when dealing with ocean environmental factors, it is difficult to accurately consider the randomness and complexity of wave loads, resulting in inaccurate calculation of hydrodynamic characteristics, which in turn affects the calculation of structural strength and fatigue life assessment. On the other hand, for the simulation of crack propagation, there is a lack of effective processing and utilization of multi-source data, and the complex relationship between crack growth and stress intensity factors cannot be fully captured, resulting in a large deviation in the calculation of crack propagation amount. Moreover, traditional methods have low calculation efficiency and are difficult to meet the requirements of rapid and accurate assessment in practical engineering.

[0004] Therefore, there is an urgent need for a method that can quickly and reliably count the fatigue life of ocean structures. Summary of the Invention

[0005] In view of this, the present invention proposes a method and system for predicting a stress intensity factor transfer function, which can efficiently and accurately count the fatigue life of ocean structures.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] Construct a model corresponding to the ocean structure, assign wet surface properties to the shell elements in contact with water outside the model, and then perform mesh division, and export the wet surface model and the overall structure model;

[0008] Input the wet surface and select the corresponding wave direction and wave frequency to calculate the corresponding hydrodynamic characteristics;

[0009] Based on the hydrodynamic characteristics, transfer the wave load to the structure and calculate the structural strength under the wave load;

[0010] Intercept the overall structure model to obtain a sub-model at the fatigue hot spot, use the structural strength of the overall structure model to map the displacement and stress of the sub-model, and construct a crack propagation model;

[0011] Insert cracks into the crack propagation area multiple times, form an extended domain of crack propagation with the initial crack position as a variable, and export the crack information in multiple calculations and the stress intensity factors corresponding to the crack information to form a data set;

[0012] After processing the data of the stress intensity factor, crack information, and wave frequency, train the multi-layer perceptron model to obtain a multi-layer perceptron model that can predict the stress intensity factor under different regular waves;

[0013] Construct a stress intensity factor transfer function based on the predicted stress intensity factor to calculate the crack growth driving force, and calculate the corresponding crack growth amount based on the crack growth driving force;

[0014] Statistically analyze the short-term sea state experience time to obtain the crack growth life curve under random sea conditions;

[0015] Based on the stress intensity factor transfer function under different crack sizes, conduct fatigue life statistics on the marine structure, and compare the statistical results with the finite element results to obtain the corresponding fatigue life of the marine structure.

[0016] On the basis of the above technical solutions, the present invention can also be improved as follows:

[0017] Optionally, inserting cracks into the crack growth area multiple times and using the initial crack position as a variable to form the expansion domain of crack growth includes:

[0018] Introduce an initial crack as the initial input;

[0019] Successively predict the stress intensity factor under regular waves with unit wave amplitude at different frequencies;

[0020] Calculate the corresponding crack growth amount according to the stress intensity factor transfer function in combination with the short-term sea state;

[0021] Judge whether the crack growth amount reaches the update threshold. If not, repeat the sea state calculation;

[0022] If it reaches, update the crack size and use the updated crack size as the new input for the next prediction cycle.

[0023] Optionally, exporting the crack information and the corresponding stress intensity factor in multiple calculations includes:

[0024] Calculate the real part equivalent stress intensity factor through formula (1);

[0025] Formula (1);

[0026] In the formula, is the real part equivalent stress intensity factor, is the actual opening mode stress intensity factor, is the actual sliding mode stress intensity factor, is the actual tearing mode stress intensity factor, is the Poisson's ratio of the material;

[0027] Calculate the imaginary part of the equivalent stress intensity factor through formula (2);

[0028] Formula (2);

[0029] In the formula, the imaginary part of the equivalent stress intensity factor, is the virtual opening mode stress intensity factor, is the virtual sliding mode stress intensity factor, is the virtual tearing mode stress intensity factor.

[0030] Optionally, the crack information in the multiple calculations and the stress intensity factor corresponding to the crack information are further included:

[0031] Calculate the stress intensity factor through formula (3);

[0032] Formula (3);

[0033] In the formula, is the stress intensity factor, is the real part of the equivalent stress intensity factor, the imaginary part of the equivalent stress intensity factor.

[0034] Optionally, the data processing of the stress intensity factor, crack information, and wave frequency includes:

[0035] Classify and label the crack information in the dataset and the stress intensity factors under different wave frequencies according to wave frequency, real part, and imaginary part;

[0036] Use the relative position of the node on the crack as the independent variable, and the coordinates of the node on the crack front and the stress intensity factor corresponding to the coordinates as the dependent variable for polynomial fitting, and evenly divide the relative position of the crack front to ensure the unity of data dimensions;

[0037] Standardize the crack information at the same crack relative position on different cracks after dimension unification to eliminate the influence of the dimension between features;

[0038] Adopt the principal component analysis method to extract the main components of the data.

[0039] Optionally, the calculation of the crack propagation driving force according to the predicted stress intensity factor includes:

[0040] Calculate the wave spectrum through formula (4);

[0041] Formula (4);

[0042] In the formula, is the wave spectrum, is the significant wave height, is the mean zero-crossing period, is the circular frequency of the wave, is the exponential function;

[0043] Calculate the stress intensity factor transfer function through formula (5);

[0044] Formula (5);

[0045] In the formula, is the stress intensity factor, is the spectral density of the stress intensity factor response, is the wave spectrum.

[0046] Optionally, calculating the corresponding crack growth amount based on the crack growth driving force includes:

[0047] Calculate the crack propagation amount through formula (6);

[0048] Formula (6);

[0049] In the formula, is the crack growth increment; is the material parameter, a proportionality coefficient reflecting the material's ability to resist fatigue crack growth, related to the material microstructure; is the duration corresponding to the short-term sea condition, and are the moments of the stress spectrum, is the stress intensity factor threshold, is the integration variable, is the material parameter, characterizing the sensitivity of the crack growth rate to the stress intensity factor amplitude ; is the stress intensity factor range under the i-th working condition probability density function.

[0050] A stress intensity factor transfer function prediction system includes:

[0051] A model derivation module for constructing a model corresponding to an offshore structure, meshing after assigning wet surface properties to the shell elements in contact with water outside the model, and deriving the wet surface model and the overall structure model;

[0052] A hydrodynamic characteristic calculation module for inputting the wet surface and selecting the corresponding wave direction and wave frequency to calculate the corresponding hydrodynamic characteristics;

[0053] A structural strength calculation module for transferring wave loads to a structure based on the hydrodynamic characteristics and calculating the structural strength under wave loads;

[0054] A crack propagation model construction module for intercepting a sub-model at a fatigue hot spot from an overall structural model, performing displacement and stress mapping on the sub-model using the structural strength of the overall structural model, and constructing a crack propagation model;

[0055] A data set construction module for inserting cracks into a crack propagation region multiple times, forming an extended domain for crack propagation with the initial crack position as a variable, and exporting crack information and the corresponding stress intensity factors in multiple calculations to form a data set;

[0056] A model training module for performing data processing on the stress intensity factors, crack information, and wave frequencies, and then training a multi-layer perceptron model to obtain a multi-layer perceptron model that can predict stress intensity factors under different regular waves;

[0057] A crack propagation amount calculation module for constructing a stress intensity factor transfer function based on the predicted stress intensity factors to calculate the crack propagation driving force, and calculating the corresponding crack propagation amount based on the crack propagation driving force;

[0058] A statistics module for statistically analyzing the experienced time of short-term sea conditions to obtain a crack propagation life curve under random sea conditions;

[0059] A fatigue life acquisition module for statistically analyzing the fatigue life of an ocean structure based on the stress intensity factor transfer function under different crack sizes, and comparing the statistical results with the finite element results to obtain the corresponding fatigue life of the ocean structure.

[0060] An electronic device includes a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, the steps of the method are implemented.

[0061] A non-transitory computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method are implemented.

[0062] The present invention has the following advantages:

[0063] In the stress intensity factor transfer function prediction method of the present invention, by endowing the shell elements in contact with water outside the model with wet surface properties, the interaction between the ocean structure and water can be accurately simulated. Combining the selected wave direction and wave frequency to calculate the hydrodynamic characteristics can fully consider the complexity and randomness of waves in the ocean environment, making the subsequent wave load transfer and structural strength calculation more in line with reality and improving the accuracy of fatigue life statistics.

[0064] In the stress intensity factor transfer function prediction method of the present invention, a crack propagation model is constructed, and the crack formation and propagation domain is inserted multiple times, comprehensively considering the crack growth process; the multi-layer perceptron model is used to predict the stress intensity factor, which can effectively process multi-source data such as crack information, stress intensity factor, and wave frequency, accurately capture the complex relationships between them, and then accurately calculate the crack propagation amount, providing a reliable basis for fatigue life statistics.

[0065] In the stress intensity factor transfer function prediction method of the present invention, the multi-layer perceptron model of machine learning is used for training and prediction. Compared with the traditional method, the calculation amount is greatly reduced, the calculation time is shortened, and the calculation efficiency is improved; at the same time, the statistical results are compared with the finite element results to ensure the reliability of the results, making the fatigue life statistics more scientific and accurate, and better guiding the design, maintenance, and safety assessment of ocean structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] For purposes of illustration and not limitation, the present invention will now be described in connection with the embodiments and drawings of the present invention, wherein:

[0067] Figure 1 is a schematic flow chart of the stress intensity factor transfer function prediction method in the embodiment of the present invention;

[0068] Figure 2 is a schematic diagram of the main components of the stress intensity factor transfer function prediction system in the embodiment of the present invention;

[0069] Figure 3 is a schematic diagram of the physical structure of the electronic device provided by the present invention;

[0070] Figure 4 is a schematic diagram of the finite element model of the cage in the embodiment of the present invention, where (a) is the structural model, indicating that the shell (shell) element is used as a whole, the internal stiffener uses the Beam (beam) element, and the partition uses the shell element; (b) is the finite element model, indicating that the shell element in contact with water on the outside of the model is assigned the wet surface property for mesh generation and the wet surface model is exported; (c) is the mass model, showing the mesh generation of the overall structure and the export of the overall structure model;

[0071] Figure 5 is a schematic diagram of the overall structural stress response in the embodiment of the present invention, where (a) is the calculation result of the Rz (reaction force) received by the structure, (b) is the overall stress distribution of the structure under the hydrostatic pressure load, (c) is the overall stress distribution of the structure under the unit amplitude wave load, and (d) is the overall stress distribution of the platform structure under the combined load;

[0072] Figure 6 is a schematic diagram of the comparison of the displacement stress nephogram of the overall model and the sub-model in the embodiment of the present invention;

[0073] Figure 7 It is a schematic diagram of the crack propagation model in the embodiment of the present invention. Among them, (a) shows the shell sub-model, and the materialized area pointed by the arrow is the position where fatigue crack propagation is likely to occur in actual engineering. (b) is the shell-solid combined model, and the arrows respectively point out the application method of the boundary conditions and the area of the shell-solid constraint. The shell-solid constraint can ensure the smooth transfer of displacement between different elements and the calculation grid is refined in this area to improve the calculation accuracy;

[0074] Figure 8 It is a schematic diagram of the displacement comparison between the sub-model and the crack propagation model in the embodiment of the present invention;

[0075] Figure 9 It is a schematic diagram of the crack front and the stress intensity factor in the embodiment of the present invention. Among them, (a) is the definition of the crack front, which defines the parameters of the inserted crack. (b) points out the area where fatigue phenomenon is likely to occur after the finite element calculation of the model. (c) is the stress intensity factor corresponding to each point of the crack front, showing the trend of the stress intensity factor at the crack front changing with the relative position;

[0076] Figure 10 It is a schematic diagram of the structure of the multi-layer perceptron in the embodiment of the present invention;

[0077] Figure 11 It is a schematic diagram of the comparison between the actual value and the predicted value of the equivalent strength factor in the embodiment of the present invention;

[0078] Figure 12 It is a schematic diagram of the comparison between the actual value and the predicted value of the equivalent stress intensity factor in the embodiment of the present invention. Detailed implementation manners

[0079] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0080] It should be noted that in the description of the present invention and the above-mentioned drawings, the terms "first", "second", etc. are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so as to implement the embodiments of the present invention described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0081] It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. The embodiments of the present invention will be described in detail below with reference to the drawings.

[0082] Figure 1 It is a schematic flow chart of the stress intensity factor transfer function prediction method in the embodiments of the present invention, as Figure 1 shown, the stress intensity factor transfer function prediction method provided by the embodiments of the present invention includes the following steps S101 to S109.

[0083] S101, construct a model corresponding to the marine structure, assign wet surface properties to the shell elements in contact with water outside the model, and then perform mesh division, and export the wet surface model and the overall structure model.

[0084] One embodiment is: model the aquaculture cage in the GENIE module of SESAM software, use shell elements as a whole, use beam elements for internal stiffeners, and use shell elements for partitions as Figure 4 shown in (a); assign wet surface properties to the shell elements in contact with water outside the model, perform mesh division, and export the wet surface model as Figure 4 shown in (b); perform mesh division on the overall structure and export the overall structure model as Figure 4 shown in (c).

[0085] S102, input the wet surface and select the corresponding wave direction and wave frequency to calculate the corresponding hydrodynamic characteristics.

[0086] Use HydroD in SESAM software to perform hydrodynamic calculations, input the wet surface of the structure, select the corresponding wave direction and wave frequency to calculate the corresponding hydrodynamic characteristics, and the hydrodynamic characteristics are expressed by the amplitude response operator (RAO).

[0087] S103, transfer the wave load to the structure based on the hydrodynamic characteristics, and calculate the structural strength under the wave load.

[0088] In the Sestra module of the SESAM software, the wave load is transferred to the structure to calculate the stress of the structure by using the calculated hydrodynamic characteristics of the structure, such as Figure 5 As shown in the figure, the cage structure adopts a single-point mooring method when working, and the structure is always in a wave-facing state. Therefore, when calculating the structural strength, only the wave force in the range of 0.1-1.8rad / s under 0-degree waves is considered, and the stress concentration area is selected as the fatigue hotspot for model refinement and subsequent crack propagation.

[0089] S104, intercepting the overall structural model to obtain a sub-model at the fatigue hotspot, using the structural strength of the overall structural model to perform displacement and stress mapping on the sub-model, and constructing a crack propagation model.

[0090] In GENIE, the overall model is intercepted to obtain the sub-model at the fatigue hotspot, and the displacement and stress mapping of the sub-model is performed using the structural strength results of the overall model, such as Figure 6 To compare the displacement and stress of the sub-model and the overall model, the results of the two are basically consistent. The crack propagation area in the sub-model is deleted and solid modeling is performed. The shell-solid coupling is used in ABAQUS to integrate them and complete the construction of the crack propagation model. Figure 7 shown.

[0091] S105, inserting cracks into the crack extension region multiple times, using the initial crack position as a variable to form an extension domain of crack extension, deriving crack information in multiple calculations and stress intensity factors corresponding to the crack information, and forming a data set.

[0092] The real part equivalent stress intensity factor is calculated by formula (1);

[0093] Formula (1);

[0094] In the formula, is the real part equivalent stress intensity factor, is the actual opening stress intensity factor, is the actual sliding stress intensity factor, is the actual tearing stress intensity factor, is the Poisson's ratio of the material;

[0095] The imaginary equivalent stress intensity factor is calculated by formula (2);

[0096] Formula (2);

[0097] In the formula, Imaginary equivalent stress intensity factor, is the virtual opening stress intensity factor, is the virtual sliding stress intensity factor, is the virtual tearing mode stress intensity factor.

[0098] Calculate the stress intensity factor through formula (3);

[0099] Formula (3);

[0100] In the formula, is the stress intensity factor, is the real part of the equivalent stress intensity factor, is the imaginary part of the equivalent stress intensity factor.

[0101] S106. After processing the data of the stress intensity factor, crack information, and wave frequency, train the multi-layer perceptron model to obtain a multi-layer perceptron model that can predict the stress intensity factor under different regular waves.

[0102] Specifically, the existing dataset information is: the crack front coordinate information and the stress intensity factor under the action of wave forces with different frequencies ; To classify the stress intensity factors corresponding to different frequencies, mark the data. The marking rule is: when the wave frequency is 0.1 rad / s, mark it as 01, mark the real part as 0, and mark the imaginary part as 1; for example, the imaginary part stress intensity factor when the frequency is 1.1 rad / s is marked as .

[0103] That is, as the crack grows, the number of grids describing the crack front increases, and the corresponding number of nodes continues to grow. Therefore, it is necessary to perform data fitting on the original data (the crack front and its corresponding stress intensity factor) to ensure dimensional unity. Use the relative position of the node on the crack as the independent variable (such as Figure 9 , point A is marked as 0, point B is marked as 1), respectively use the coordinates of the nodes on the crack front and their corresponding stress intensity factors as the dependent variables for polynomial fitting, evenly divide the relative position of the crack front (from 0 to 1, with an interval of 0.02), and obtain the corresponding crack information.

[0104] After dimensional unity, perform standardization processing on the crack information at the same crack relative position on different cracks. Standardization processing in machine learning is used to scale features to the same range or scale, eliminating the influence of the dimension between features. This helps to accelerate the convergence of the optimization algorithm, improve the accuracy and stability of the model, and ensure the balanced contribution of different features to the model. Through standardization, the model training process becomes more efficient and reduces the instability in numerical calculations.

[0105] In machine learning, the input dimension determines the richness of features and the complexity of the model, while the output dimension directly affects the model's predictive ability and the type of task. A too-high input dimension may lead to increased computational complexity and overfitting, while the output dimension needs to match the task requirements to ensure that the model can effectively perform classification or regression. To reduce the input and output dimensions while retaining the main information therein and ensuring the robustness of the model, principal component analysis (PCA) is selected to extract the principal components of the data. Principal component analysis aims to transform the data into a new coordinate system, retain the main variability in the data, and at the same time reduce the input and output dimensions. PCA simplifies the data structure by identifying the principal components in the data, that is, the linear combinations in the directions of the largest variances in the data. It projects the data points in the original feature space onto these principal components, thereby reducing the number of features while retaining as much of the main information of the data as possible.

[0106] Therefore, in the present invention, it is regarded as a dimensionality reduction method, which not only helps to reduce the computational complexity, but also reduces the risk of overfitting and improves the robustness of the model. In this way, PCA makes data processing more efficient and provides a more concise and effective feature set for subsequent model training and analysis.

[0107] Machine learning is an important branch of artificial intelligence that enables a computer to automatically learn and improve through data and experience without explicit programming. Through algorithms and models, machine learning can identify the internal relationships in the data, make predictions, and make decisions. It is mainly divided into types such as supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.

[0108] As Figure 10 shown, the multi-layer perceptron (MLP) is a basic feedforward neural network composed of multiple layers, including an input layer, hidden layers, and an output layer. Each neuron processes the input through an activation function and can capture complex patterns and relationships. The MLP calculates the output through forward propagation and uses the backpropagation algorithm to update the network weights to minimize the prediction error. This model is widely used in tasks such as classification, regression, and image recognition and is one of the basic architectures of deep learning.

[0109] The same crack will produce different stress responses under the action of waves with different frequencies, which leads to different stress intensity factors. Therefore, it is necessary to separately train the surrogate models under a series of regular waves. The training data structures under different sea conditions are similar but still have different differences. The stress intensity factors under the same sea condition , and The three types of data structures are also different and require different MLP structures (number of hidden layers, activation function, number of neurons, batch size, number of iterations, etc.) to optimize the prediction ability of the surrogate model. To achieve this goal, the present invention uses grid search during model training. Grid search is a hyperparameter optimization technique that systematically traverses all possible hyperparameter combinations to find the best configuration. It combines the predefined hyperparameter value ranges into a grid, trains the model one by one, and evaluates the performance to determine the optimal parameter settings. Table 1 shows the parameter settings.

[0110] Table 1

[0111]

[0112] Mean Squared Error (MSE) is a commonly used metric to evaluate the performance of a regression model. It quantifies the accuracy of the model by calculating the average of the squares of the differences between the predicted values and the actual values. Taking the surrogate model with a wave frequency of 0.1 rad / s as an example, Table 2 shows its MLP structure;

[0113] Table 2

[0114]

[0115] The learning rate is set to 0.001. To avoid overfitting caused by too many iterations, 'early stopping' is set during model training. When the model loss value no longer decreases continuously, it is considered that the model has converged, and at this time, the model training is stopped. The training iteration times of different models are mostly distributed between 200 and 300 times, and the training time of a single model is about 30 seconds.

[0116] S107, construct a stress intensity factor transfer function according to the predicted stress intensity factor to calculate the crack growth driving force, and calculate the corresponding crack growth amount based on the crack growth driving force.

[0117] Calculate the wave spectrum through formula (4);

[0118] Formula (4);

[0119] In the formula, is the wave spectrum, is the significant wave height, is the average zero-crossing period, is the circular frequency of the wave, is the exponential function;

[0120] Calculate the stress intensity factor transfer function through formula (5);

[0121] Formula (5);

[0122] In the formula, is the stress intensity factor, is the stress intensity factor response spectral density, is the wave spectrum.

[0123] Derive the boundary node displacements of the sub-model under different wave loads, and apply them to the crack propagation model in ABAQUS (such as Figure 7 ), and finally act on the solid part through the transmission of the shell. As Figure 8 , it is the stress comparison between the sub-model and the crack propagation model.

[0124] To achieve fatigue crack propagation under random wave loads, use the spectral analysis method. By constructing the stress intensity factor transfer function and combining the random wave as the crack propagation driving force, the crack growth is realized.

[0125] After obtaining the crack driving force, the corresponding crack propagation amount needs to be calculated. Since the wave can be regarded as a narrow-band Gaussian process, the stress intensity factor response on this linear system of offshore structures can also be regarded as a narrow-band Gaussian process, and it is considered that the distribution of the stress intensity factor amplitude satisfies the Rayleigh distribution. Then, its probability density function is:

[0126] ;

[0127] And the stress intensity factor range is twice the stress intensity factor amplitude, that is, , then the probability density function of the stress intensity factor range is:

[0128] ;

[0129] Among them, the Rayleigh distribution parameter is expressed as:

[0130] ;

[0131] And its average zero-crossing rate is expressed as:

[0132] ;

[0133] Combined with the classical stable stage crack propagation model Paris formula:

[0134] ;

[0135] Among them, is the material parameter, which is the proportional coefficient reflecting the material's resistance to fatigue crack propagation and is related to the material's microstructure, is the material parameter, which characterizes the crack propagation rate corresponding to the stress intensity factor amplitude The sensitivity is the crack growth rate, and

[0136] the crack growth increment under a single short-term sea condition can be calculated according to the following formula:

[0137] ;

[0138] ;

[0139] wherein, is the number of cycles under the th short-term sea condition, is the duration corresponding to the short-term sea condition, and

[0140] is the average zero-crossing rate under the corresponding sea condition.

[0141] ;

[0142] Once the crack growth increment is obtained, the forward growth of the crack and the calculation of the life can be carried out.

[0143] The present invention develops an automatic crack growth program using the PYTHON language, and realizes the automatic crack growth of offshore structures based on spectral analysis by connecting the stress intensity prediction model and the crack growth model. The main steps of the automatic crack growth program are as follows:

[0144] Introduce an initial crack as the initial input;

[0145] Successively predict the stress intensity factors under regular waves with unit wave amplitude at different frequencies;

[0146] Construct a transfer function according to the predicted stress intensity factors;

[0147] Calculate the crack growth amount under this sea condition according to the transfer function in combination with the short-term sea condition;

[0148] Judge whether the crack growth amount reaches the update threshold. If it does not reach the threshold, repeat the calculation for the sea condition;

[0149] If the update threshold is reached, update the crack size and use it as the input for the next prediction.

[0150] S108. Statistically obtain the crack growth life curve under the random sea condition by counting the experienced time of the short-term sea condition.

[0151] S109. Based on the stress intensity factor transfer function under different crack sizes, the fatigue life of the offshore structure is statistically analyzed, and the statistical results are compared with the finite element results to obtain the corresponding fatigue life of the offshore structure.

[0152] Under each short-term sea condition, the structure is affected by waves and the stress intensity factor is recalculated after crack propagation, which makes the calculation results more accurate. However, during the service life of the offshore structure, it will be affected by more than 10^4 short-term sea conditions. This means that if the finite element method is used, a large number of calculations will be carried out for crack propagation after each short-term sea condition, seriously affecting the calculation efficiency. Therefore, a crack update threshold was set in the previous method: when the crack propagation amount reaches the specified threshold, the crack is updated and the stress intensity factor is calculated. Although this operation improves the calculation efficiency, it reduces the calculation accuracy. Using the machine learning method, the time-consuming and laborious finite element calculation is replaced by the quickly predicted MLP model. Therefore, a smaller threshold can be used for crack update, and this method not only improves the calculation efficiency but also greatly improves the calculation accuracy.

[0153] The coefficient of determination (R², R-squared) is an index to measure the goodness of fit of the regression model, indicating the explanatory ability of the independent variable for the dependent variable. Its value ranges from 0 to 1, and the closer it is to 1, the stronger the explanatory ability of the model for the data. The data is trained separately according to the classification in part (a), and 1 - R² is used as the evaluation index of the model. Table 3 shows the evaluation values of the surrogate models corresponding to 24 sea conditions.

[0154] Table 3

[0155]

[0156] After PCA, the stress intensity factor is converted into 3 principal components. Accurately predicting the principal components can obtain the correct stress intensity factor, and the stress intensity factor, as the basis for constructing the transfer function, determines the final fatigue life statistics.

[0157] Next, the accuracy of the surrogate model will be verified by comparing the prediction of the transfer function.

[0158] The prediction process of the SIF spectrum under random load conditions using the machine learning method is divided into 3 parts:

[0159] Specify the crack front, frequency, and wave direction;

[0160] The machine learning model predicts the stress intensity factor under the corresponding working conditions;

[0161] Extract the stress intensity factor of the real and imaginary parts along the depth and length directions of the lower edge;

[0162] The stress intensity factors at different frequencies are combined to form a stress intensity factor transfer function;

[0163] Combined with the cyclic scatter diagram, the Rayleigh distribution of the SIF range is obtained and crack propagation is carried out;

[0164] The crack size is updated and this cycle is repeated.

[0165] Figure 11 It is a comparison schematic diagram of the actual value and predicted value of the equivalent strength factor. Figure 12 It is a comparison diagram of the actual value and predicted value of the equivalent stress intensity factor. The prediction results of the stress intensity factors for different cracks at different wave directions and frequencies are good, indicating that the SIF spectrum prediction method based on machine learning under this process framework is excellent; 80×3×23×2 = 11040 groups of data are cut, and the data of the last 20 expansion steps, that is, the data of large cracks, are intercepted for testing the extrapolation ability of the model to predict the stress intensity factor; the MLP model parameter configuration and training process are the same as above, and the large crack data are used as input and put into the model for prediction. As Figure 12 shown, the overall error rate does not exceed 3%.

[0166] The random load condition is realized by changing the combination of two variables, the wave direction angle and the wave frequency. The crack update threshold is set to 0.05 mm, the course angle ranges from 0° to 60°, and the value is taken every 30°; the wave frequency is 0.2~2.4 rad / s, and the value is taken every 0.1 rad / s; the SIF spectra under different wave directions are predicted respectively. Given the crack front parameters, the trained machine learning model is used to predict the stress intensity factor. The short-term sea condition is input and the crack propagation amount is calculated by combining with the Paris formula. When the cumulative expansion amount reaches the update threshold, the crack front is updated to establish the SIF spectra under each size and each random load condition.

[0167] Among them, when the wave direction angle is 0°, the overall SIF is relatively small, indicating that when the wave direction angle is 0°, the direction of the first principal stress is close to the crack direction, and the crack propagation is slower. When the wave direction angles are 30° and 60°, the SIF values are larger and the crack propagation rate is faster. When the wave direction angles are 0° and 30°, in the later stage of crack propagation, the SIF becomes steeper and its growth ratio gradually becomes larger. This is because as the crack length increases, the stress concentration phenomenon at the crack tip is more significant. For the wave direction angle of 60°, the change of SIF in the later stage gradually flattens. This may be because under this wave condition, the main contribution to crack growth is not the mode I stress intensity factor. Therefore, the crack is simplified to a two-dimensional shape, and only the opening-type crack is considered, which will lead to different from the real development law, deviate from the original crack growth path, and the growth of the stress intensity factor will gradually slow down.

[0168] The method based on fracture mechanics needs to calculate the crack growth amount under short-term sea conditions. According to the aforementioned theory, the distribution of SIF needs to be calculated first. Regarding the wave as a narrow-band Gaussian process, the stress intensity factor range under linear transformation satisfies the Rayleigh distribution. Combining with the short-term sea conditions, the corresponding probability density function of the SIF distribution is obtained; the sea conditions with a wave direction angle of 30 degrees, Hs = 2.5m, and Tz = 8.5, and the probability density functions corresponding to different crack depths at the same crack position. As the crack grows, the probability density gradually distributes to the region with a larger SIF, and the overall SIF amplitude level increases.

[0169] According to the probability density function of the SIF amplitude and combined with the Paris formula, the crack growth amount under short-term sea conditions can be calculated. At the same time, by statistically analyzing the experienced time of short-term sea conditions, the crack growth life curve under random sea conditions can be obtained. Based on the stress transfer function under different crack sizes, the fatigue life of the structure is statistically analyzed and compared with the finite element results. The calculation time of the life statistics predicted by machine learning is at the minute level, which greatly reduces the time cost. Compared with the finite element results, the life is more conservative, and the error is -6.2%.

[0170] Figure 2 It is a schematic diagram of the main components of the stress intensity factor transfer function prediction system in the embodiment of the present invention. As Figure 2 shown, the stress intensity factor transfer function prediction system 1 provided by the embodiment of the present invention includes a model export module 10, a hydrodynamic characteristic calculation module 20, a structural strength calculation module 30, a crack growth model construction module 40, a data set construction module 50, a model training module 60, a crack growth amount calculation module 110, a statistics module 80, and a fatigue life acquisition module 90.

[0171] The model export module 10 is used to construct a model corresponding to the marine structure, assign wet surface properties to the shell elements in contact with water outside the model, perform mesh division, and export the wet surface model and the overall structure model;

[0172] The hydrodynamic characteristic calculation module 20 is used to input the wet surface and select the corresponding wave direction and wave frequency to calculate the corresponding hydrodynamic characteristics;

[0173] The structural strength calculation module 30 is used to transfer the wave load to the structure based on the hydrodynamic characteristics and calculate the structural strength under the wave load;

[0174] The crack growth model construction module 40 is used to intercept the overall structure model to obtain a sub-model at the fatigue hot spot, use the structural strength of the overall structure model to map the displacement and stress of the sub-model, and construct a crack growth model;

[0175] The dataset construction module 50 is configured to insert cracks into the crack propagation area multiple times, form an extended domain of crack propagation with the initial crack position as a variable, export the crack information in multiple calculations and the stress intensity factors corresponding to the crack information, and form a dataset;

[0176] The model training module 60 is configured to perform data processing on the stress intensity factors, crack information, and wave frequencies, and then train a multi-layer perceptron model to obtain a multi-layer perceptron model that can predict the stress intensity factors under different regular waves;

[0177] The crack propagation amount calculation module 70 is configured to construct a stress intensity factor transfer function based on the predicted stress intensity factors to calculate the crack propagation driving force, and calculate the corresponding crack propagation amount based on the crack propagation driving force;

[0178] The statistics module 80 is configured to count the short-term sea condition experience time to obtain the crack propagation life curve under random sea conditions;

[0179] The fatigue life acquisition module 90 is configured to perform fatigue life statistics on the ocean structure based on the stress intensity factor transfer function under different crack sizes, and compare the statistical results with the finite element results to obtain the corresponding fatigue life of the ocean structure.

[0180] Figure 3 The following is a schematic diagram of the physical structure of the electronic device provided by the embodiment of the present invention. As Figure 3 shown, the electronic device 110 includes: a processor 1101 (processor), a memory 1102 (memory), and a bus 1103;

[0181] Among them, the processor 1101 and the memory 1102 communicate with each other through the bus 1103;

[0182] The processor 1101 is configured to call the program instructions in the memory 1102 to execute the methods provided in the above method embodiments to execute the methods provided in the embodiments of the present invention.

[0183] This embodiment provides a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the methods provided in the embodiments of the present invention.

[0184] Those of ordinary skill in the art can understand that all or part of the steps to implement the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various storage media such as ROM, RAM, magnetic disk, or optical disk that can store program codes.

[0185] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub - combinations and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A stress intensity factor transfer function prediction method, characterized in that: include: Construct a model corresponding to the marine structure, assign wet surface properties to the shell elements outside the model that are in contact with water, and then perform meshing, and derive the wet surface model and the overall structure model; Input the wet surface and select the corresponding wave direction and wave frequency to calculate the corresponding hydrodynamic characteristics; transferring the wave load to the structure based on the hydrodynamic characteristics, and calculating the structural strength under the wave load; The overall structural model is intercepted to obtain a sub-model at the fatigue hotspot, and the structural strength of the overall structural model is used to perform displacement and stress mapping on the sub-model to construct a crack propagation model; Insert cracks into the crack extension region multiple times, use the initial crack position as a variable to form an extension domain of crack extension, derive crack information in multiple calculations and stress intensity factors corresponding to the crack information, and form a data set; After data processing of the stress intensity factor, crack information and wave frequency, a multilayer perceptron model is trained to obtain a multilayer perceptron model capable of predicting stress intensity factors under different regular waves; constructing a stress intensity factor transfer function according to the predicted stress intensity factor to calculate the crack propagation driving force, and calculating the corresponding crack propagation amount based on the crack propagation driving force; The crack growth life curve under random sea conditions is obtained by statistically analyzing the short-term sea condition experience time; The fatigue life of marine structures is statistically analyzed based on the stress intensity factor transfer function under different crack sizes, and the statistical results are compared with the finite element results to obtain the corresponding fatigue life of the marine structures; The method of calculating the crack propagation driving force by constructing a stress intensity factor transfer function according to the predicted stress intensity factor comprises: The wave spectrum is calculated by formula (4); Formula (4): In the formula, For the wave spectrum, is the effective wave height, is the average zero-crossing period, is the circular frequency of the wave, is an exponential function; The stress intensity factor transfer function is calculated by formula (5); Formula (5): In the formula, is the stress intensity factor, is the stress intensity factor response spectral density, For the wave spectrum.

2. The stress intensity factor transfer function prediction method according to claim 1, characterized in that: The method of inserting cracks into the crack extension region multiple times and forming an extension domain of crack extension with the initial crack position as a variable includes: Introduce an initial crack as initial input; Successively predict the stress intensity factor under the action of regular waves of unit amplitude of different frequencies; The corresponding crack extension is calculated based on the stress intensity factor transfer function combined with the short-term sea conditions; determining whether the crack extension reaches an update threshold, and if not, repeating the sea condition calculation; If it is reached, the crack size is updated and the updated crack size is used as the new input for the next prediction cycle.

3. The stress intensity factor transfer function prediction method according to claim 1, characterized in that: The exporting of the crack information in the multiple calculations and the stress intensity factor corresponding to the crack information includes: The real part equivalent stress intensity factor is calculated by formula (1); Formula (1): In the formula, is the real part equivalent stress intensity factor, is the actual opening stress intensity factor, is the actual sliding stress intensity factor, is the actual tearing stress intensity factor, is the Poisson's ratio of the material; The imaginary equivalent stress intensity factor is calculated by formula (2); Formula (2): In the formula, Imaginary equivalent stress intensity factor, is the virtual opening stress intensity factor, is the virtual sliding stress intensity factor, is the virtual tearing stress intensity factor.

4. The stress intensity factor transfer function prediction method according to claim 3, characterized in that: The method of deriving the crack information in the multiple calculations and the stress intensity factor corresponding to the crack information further includes: The stress intensity factor is calculated by formula (3); Formula (3): In the formula, is the stress intensity factor, is the real part equivalent stress intensity factor, Imaginary equivalent stress intensity factor.

5. The stress intensity factor transfer function prediction method according to claim 1, characterized in that: The data processing of the stress intensity factor, crack information and wave frequency includes: The crack information and stress intensity factors under different wave frequencies in the data set are classified and marked according to the wave frequency, real part and imaginary part; The relative position of the node on the crack is taken as the independent variable, and the coordinates of the node on the crack front and the stress intensity factor corresponding to the coordinates are taken as the dependent variables for polynomial fitting. The relative position of the crack front is evenly divided to ensure the uniformity of the data dimension. The crack information at the relative position of the same crack on different cracks after the dimensions are unified is standardized to eliminate the dimensional influence between the features; The principal component analysis method is used to extract the principal components of the data.

6. The stress intensity factor transfer function prediction method according to claim 1, characterized in that: The calculating the corresponding crack extension amount based on the crack extension driving force comprises: The crack propagation is calculated by formula (6); Formula (6): In the formula, is the crack extension increment; It is a material parameter, which reflects the proportional coefficient of the material's ability to resist fatigue crack growth and is related to the material's microstructure; is the duration corresponding to the short-term sea conditions, and is the moment of the stress spectrum, is the stress intensity factor threshold value, is the integration variable, is a material parameter that characterizes the effect of crack growth rate on stress intensity factor amplitude. sensitivity; is the range of stress intensity factor under the ith working condition The probability density function of .

7. A system for predicting a stress intensity factor transfer function, characterized in that: include: Model export module, used to build the model corresponding to the marine structure, assign wet surface properties to the shell elements outside the model that contact the water, perform meshing, and export the wet surface model and the overall structure model; A hydrodynamic characteristic calculation module is used to input a wet surface and select a corresponding wave direction and wave frequency to calculate the corresponding hydrodynamic characteristics; a structural strength calculation module, used to transfer the wave load to the structure based on the hydrodynamic characteristics and calculate the structural strength under the wave load; The crack growth model construction module is used to intercept the overall structural model to obtain the sub-model at the fatigue hotspot, and use the structural strength of the overall structural model to perform displacement and stress mapping on the sub-model to construct the crack growth model; A data set construction module is used to insert cracks into the crack extension area multiple times, use the initial crack position as a variable to form an extension domain of crack extension, derive crack information in multiple calculations and stress intensity factors corresponding to the crack information, and form a data set; A model training module is used to train a multilayer perceptron model after data processing of the stress intensity factor, crack information and wave frequency, so as to obtain a multilayer perceptron model capable of predicting stress intensity factors under different regular waves; A crack extension amount calculation module, used to construct a stress intensity factor transfer function according to the predicted stress intensity factor to calculate the crack extension driving force, and calculate the corresponding crack extension amount based on the crack extension driving force; The statistical module is used to calculate the short-term sea condition experience time to obtain the crack growth life curve under random sea conditions; The fatigue life acquisition module is used to perform fatigue life statistics on marine structures based on the stress intensity factor transfer function under different crack sizes, and compare the statistical results with the finite element results to obtain the corresponding fatigue life of the marine structure.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A non-transitory computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images