Surface crack stress intensity factor prediction method and device based on machine learning

Abstract

The invention relates to the technical field of structural engineering, and provides a surface crack stress intensity factor prediction method and device based on machine learning, and the method comprises the steps: building a finite element simulation model containing a surface crack, calculating the stress intensity factor of the front edge of the surface crack, verifying through an empirical formula, and generating a finite element simulation data set; performing dimensionless processing on the size of the surface crack, the size of the load bearing structure and the stress intensity factor of the front edge of the surface crack in the finite element simulation data set to generate a machine learning data set; optimizing the back propagation neural network through a hybrid optimization algorithm, and constructing a stress intensity factor prediction model based on machine learning; adjusting hyper-parameters of the neural network, and verifying a fitting effect of the stress intensity factor prediction model based on machine learning; based on the stress intensity factor prediction model, establishing a visual interaction interface for related parameter input, image output, prediction result output and comparison result output; while the calculation efficiency is maintained, the calculation precision and the application range are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of structural engineering, and in particular, to a method and device for predicting the stress intensity factor of surface cracks based on machine learning. Background Art

[0002] Fatigue and fracture are one of the main causes of engineering structure failures. Scientifically evaluating the reliability of structures with crack defects is of great significance in ensuring the safety and economy of production systems. Among various engineering equipment, semi-elliptical cracks on the surface of pressure vessels are a common form of defect. Such cracks are usually formed under the action of stress concentration caused by rolling and welding processes, long-term fatigue damage, and corrosion, etc., and may gradually expand under continuous stress, eventually leading to equipment failure or even catastrophic accidents. The stress intensity factor at the crack tip is one of the core parameters for measuring the fracture behavior of materials and is also the basis for evaluating the stability and safety of cracked structures. Quickly and accurately obtaining the stress intensity factor value is crucial for the safety assessment of engineering equipment.

[0003] Currently, for the calculation of the stress intensity factor of surface cracks, the invention patent with the patent publication number CN114492110B discloses a method and system for calculating the stress intensity factor of surface cracks on a disk based on weight functions. This method uses the stress intensity factor obtained by the empirical formula Newman-Raju as a reference solution to solve the weight coefficients in the weight function formula. The stress intensity factor of the surface crack can be obtained from the weight function and the stress distribution. However, the applicable range of the Newman-Raju formula is limited and there are errors in the reference solution itself. The weight function will inherit this deviation, thus affecting the reliability of subsequent calculations. Another method is to calculate the fitting coefficient and the crack shape coefficient by fitting the stress distribution perpendicular to the crack plane, and solve the stress intensity factor without relying on finite element analysis. The above technologies have at least the following problems: They are only applicable to semi-elliptical surface cracks under single-position and single-load conditions, and the calculation accuracy is limited by fitting and formulas, unable to cover the changes in crack position, direction, and load type in actual working conditions; the weight function method is limited by the accuracy and application range of the empirical formula, and a two-dimensional integral operation of the product of the weight function and the stress distribution is required to obtain the stress intensity factor. The fitting function method is only applicable to surface cracks under single-position and single-load conditions, reducing the applicability and calculation accuracy, and also reducing the calculation efficiency.

[0004] Therefore, how to improve the calculation efficiency, applicability, and calculation accuracy of the stress intensity factor of surface cracks in the face of non-standard crack geometries or complex load conditions has become an urgent technical problem to be solved. Summary of the Invention

[0005] In view of this, the main object of the present invention is to provide a method and device for predicting the stress intensity factor of surface cracks based on machine learning, aiming to solve the technical problems of improving the calculation efficiency, applicability and calculation accuracy of the stress intensity factor of surface cracks.

[0006] To achieve the above object, in a first aspect, the present invention provides a method for predicting the stress intensity factor of surface cracks based on machine learning, including the following steps:

[0007] S1, establish a finite element simulation model containing surface cracks, calculate the stress intensity factor at the front edge of the surface crack, verify the stress intensity factor of the finite element simulation model through the Newman-Raju empirical formula, and generate a finite element simulation data set;

[0008] S2, perform dimensionless processing on the size of the surface crack, the size of the load-bearing structure, and the stress intensity factor at the front edge of the surface crack in the finite element simulation data set to generate a machine learning data set;

[0009] S3, optimize the backpropagation neural network through a hybrid optimization algorithm to construct a stress intensity factor prediction model based on machine learning;

[0010] S4, adjust the neural network hyperparameters, and verify the fitting effect of the stress intensity factor prediction model based on machine learning according to the evaluation index;

[0011] S5, based on the stress intensity factor prediction model based on machine learning, establish a visual interaction interface for input of crack parameters, input of load parameters, input of neural network hyperparameters, output of the image of the load-bearing structure containing surface cracks, output of prediction results, and output of comparison results of empirical formulas.

[0012] Optionally, the establishment of the finite element simulation model containing surface cracks includes the following steps:

[0013] S11, establish a finite element model of the load-bearing structure through ABAQUS software, and export the output file of the finite element model of the load-bearing structure;

[0014] S12, insert a semi-elliptical surface crack into the model of the output file of the load-bearing structure model through FRANC3D software, and output the finite element model of the load-bearing structure containing surface cracks.

[0015] Optionally, the insertion of a semi-elliptical surface crack into the model of the output file of the load-bearing structure model through FRANC3D software includes the following steps:

[0016] S121, divide the finite element model of the load-bearing structure into an overall model and a local model;

[0017] S122. Define the minor axis dimension and major axis dimension of the semi-elliptical surface crack, and introduce the semi-elliptical surface crack into the local model to form a local model with a surface crack.

[0018] S123. Mesh the local model with a surface crack.

[0019] S124. Merge the global model and the local model with a surface crack to form a finite element model of the load-bearing structure with a surface crack.

[0020] Optionally, the dimensionless processing of the size of the surface crack, the size of the load-bearing structure, and the stress intensity factor at the front of the surface crack in the finite element simulation dataset to generate the finite element simulation dataset includes the following steps:

[0021] S21. Perform dimensionless processing on the minor axis dimension of the crack, the major axis dimension of the crack, the thickness of the load-bearing structure, the inner radius of the load-bearing structure, and the stress intensity factor, and use the normalized distance at the crack front to replace the parametric angle of the ellipse to form dimensionless quantities. The dimensionless quantities are specifically: the ratio of the minor axis of the crack to the major axis of the crack, the ratio of the minor axis of the crack to the thickness of the load-bearing structure, the ratio of the thickness of the load-bearing structure to the characteristic size of the load-bearing structure, the normalized distance at the crack front, and the surface crack boundary correction factor.

[0022] S22. Generate an axial semi-elliptical crack dataset and a circumferential semi-elliptical crack dataset.

[0023] Optionally, the optimization of the backpropagation neural network by the hybrid optimization algorithm includes the following steps:

[0024] S31. Optimize the weights and biases of the backpropagation neural network through the particle swarm algorithm to obtain the optimal combination of weights and biases of the backpropagation neural network.

[0025] S32. Apply the crossover and mutation operations of the genetic algorithm to the particle swarm algorithm to enhance the population diversity and global search ability.

[0026] Optionally, the step S1 is:

[0027] S11a. Based on the position of the surface crack in the load-bearing structure, generate a finite element simulation model with surface cracks at different positions.

[0028] Among them, the position where the surface crack is located in the load-bearing structure is specifically the axial direction of the outer surface of the load-bearing structure, the axial direction of the inner surface of the load-bearing structure, the circumferential direction of the outer surface of the load-bearing structure, and the circumferential direction of the inner surface of the load-bearing structure.

[0029] Among them, the parameter range includes circular cracks to long elliptical cracks, shallow cracks to deep cracks, and thin-walled load-bearing structures to thick-walled load-bearing structures.

[0030] S12a. The finite element simulation model with surface cracks includes a load-bearing structure model with surface cracks under internal pressure load and uniform tensile load at both ends.

[0031] Optionally, the evaluation criteria in step S4 include at least one of coefficient of determination, mean absolute error, mean bias error, and mean absolute percentage error.

[0032] In a second aspect, the present invention provides a device for predicting stress intensity factor of surface cracks based on machine learning. The device includes:

[0033] A data set acquisition module, configured to receive a finite element simulation model with surface cracks established by a user, calculate the stress intensity factor at the front edge of the surface crack, verify the stress intensity factor of the finite element simulation model through the Newman-Raju empirical formula, and generate a finite element simulation data set;

[0034] A data preprocessing module, configured to perform dimensionless processing on the size of the surface crack, the size of the load-bearing structure, and the stress intensity factor at the front edge of the surface crack in the finite element simulation data set to generate a machine learning data set;

[0035] A machine learning model generation module, configured to optimize the backpropagation neural network through a hybrid optimization algorithm to construct a stress intensity factor prediction model based on machine learning;

[0036] A model training and verification module, configured to adjust the hyperparameters of the neural network and verify the fitting effect predicted by the stress intensity factor prediction model based on machine learning according to the evaluation index;

[0037] A graphical user interface output module, configured to establish a visual interaction interface for input of crack parameters, input of load parameters, input of neural network hyperparameters, output of an image of the load-bearing structure with surface cracks, output of prediction results, and output of comparison results of empirical formulas based on the stress intensity factor prediction model of machine learning.

[0038] Optionally, the establishment of the finite element simulation model with surface cracks includes:

[0039] A load-bearing structure modeling module, configured to establish a finite element model of the load-bearing structure through ABAQUS software and export the output file of the finite element model of the load-bearing structure;

[0040] A crack insertion modeling module, configured to insert a semi-elliptical surface crack into the model of the output file of the load-bearing structure model through FRANC3D software and output the finite element model of the load-bearing structure with surface cracks.

[0041] In a third aspect, the present invention provides an electronic device, including a memory and a processor, wherein:

[0042] The memory is used to store a computer program;

[0043] The processor is configured to execute the computer program to implement the surface crack stress intensity factor prediction method based on machine learning disclosed in the first aspect above.

[0044] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and the computer program is used to be run by a processor to implement the surface crack stress intensity factor prediction method based on machine learning disclosed in the first aspect above.

[0045] In a fifth aspect, the present invention provides a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the surface crack stress intensity factor prediction method based on machine learning disclosed in the first aspect above is implemented.

[0046] In the technical solution provided by the present invention, first, a finite element model of a load-bearing structure model with a surface crack covering different positions, sizes, under internal pressure load and under uniform tensile load at both ends of a surface semi-elliptical crack is established in advance, and a machine learning data set is formed by a neural network model learning the finite element model data set, and a stress intensity factor prediction model based on machine learning is formed. When a user inputs the size parameters of a crack through a graphical visualization interface, the prediction model can be directly called to intuitively obtain the predicted surface crack stress intensity factor, avoiding the process of finite element modeling, mesh generation and calculation when the user uses it, avoiding the long time-consuming of numerical simulation and the result being easily affected by the mesh quality, and improving the calculation efficiency of the surface crack stress intensity factor;

[0047] Second, the positions of the surface semi-elliptical crack on the load-bearing structure in the present invention include the outer surface axial direction, the inner surface axial direction, the outer surface circumferential direction and the inner surface circumferential direction, and the parameter range includes circular cracks to long elliptical cracks, shallow cracks to deep cracks, thin-walled load-bearing structures to thick-walled load-bearing structures; the established finite element model includes a load-bearing structure model with a surface crack under internal pressure load and under uniform tensile load at both ends; basically covering the prediction of the surface crack stress intensity factor under non-standard crack geometries or complex load conditions, improving the applicability and accuracy of the calculation of the surface crack stress intensity factor;

[0048] Third, the machine learning training data set is generated by the joint simulation calculation of the finite element software ABAQUS and the crack analysis software FRANC3D, avoiding the problems of insufficient accuracy and limited applicable range of using empirical formulas, and can realize the prediction of the stress intensity factor of surface cracks with different sizes and positions under complex loads. Description of the Drawings

[0049] One or more embodiments are exemplarily illustrated by corresponding drawings. These exemplary illustrations do not limit the embodiments. Elements with the same reference numerals in the drawings represent similar elements. Unless otherwise stated, the drawings in the figures do not constitute a scale limitation.

[0050] Figure 1 It is a schematic flowchart of a method for predicting the stress intensity factor of surface cracks based on machine learning disclosed in one or more embodiments of the present application;

[0051] Figure 2 It is a schematic diagram of modules of a device for predicting the stress intensity factor of surface cracks based on machine learning disclosed in one or more embodiments of the present application.

[0052] Figure 3 It is a schematic diagram of modules for establishing a finite element simulation model with surface cracks disclosed in one or more embodiments of the present application;

[0053] Figure 4 It is a schematic structural diagram of a structure with a surface semi-elliptical crack under load disclosed in one or more embodiments of the present application;

[0054] Figure 5 It is a schematic structural diagram of a finite element model of a load-bearing structure with surface cracks disclosed in one or more embodiments of the present application;

[0055] Figure 6 It is a schematic diagram of the regression value between the predicted value and the simulated value of the PSOGA-BP neural network model disclosed in one or more embodiments of the present application;

[0056] Figure 7 It is a schematic diagram of a graphical user interface disclosed in one or more embodiments of the present application;

[0057] In the figure: 100 - Dataset acquisition module; 110 - Load-bearing structure modeling module; 120 - Crack insertion modeling module; 200 - Data preprocessing module; 300 - Machine learning model generation module; 400 - Model training and verification module; 500 - Graphical user interface output module. Detailed Description of the Embodiments

[0058] For the convenience of understanding the present invention, the present invention will be described in more detail below with reference to the accompanying drawings and specific embodiments. It should be noted that when an element is described as "fixed to" another element, it can be directly on the other element, or there can be one or more intermediate elements therebetween. When an element is described as "connected to" another element, it can be directly connected to the other element, or there can be one or more intermediate elements therebetween. The terms "vertical", "horizontal", "left", "right", "inside", "outside" and similar expressions used in this specification are only for the purpose of illustration. In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating relative importance or implicitly indicating the quantity of the indicated technical features. Thus, unless otherwise specified, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features; the meaning of "a plurality" is two or more. The term "comprising" and any variation thereof means non-exclusive inclusion, and there may be or be added one or more other features, integers, steps, operations, units, components and / or their combinations.

[0059] In addition, unless otherwise clearly specified and defined, the terms "installed", "connected", "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, or the communication inside two elements. All the technical and scientific terms used in this specification have the same meaning as those commonly understood by those skilled in the technical field to which the present invention belongs. The terms used in the description of the present invention are only for the purpose of describing specific embodiments and are not used to limit the present invention. The term "and / or" used in this specification includes any and all combinations of one or more of the related listed items.

[0060] In addition, the technical features involved in different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0061] Embodiment 1

[0062] To achieve the above object, an embodiment of the present invention provides a method for predicting the stress intensity factor of surface cracks based on machine learning. Please refer to the attached Figures 1 - 7 , where the attached Figure 1 is a schematic flow chart of a method for predicting the stress intensity factor of surface cracks based on machine learning. Specifically, it includes the following steps:

[0063] S1, establish a finite element simulation model containing surface cracks, calculate the stress intensity factor at the front edge of the surface crack, verify the stress intensity factor of the finite element simulation model through the Newman-Raju empirical formula, and generate a finite element simulation data set;

[0064] S2. Nondimensionally process the size of the surface crack, the size of the load-bearing structure, and the stress intensity factor at the front of the surface crack in the finite element simulation dataset to generate a machine learning dataset;

[0065] S3. Optimize the backpropagation neural network through a hybrid optimization algorithm to construct a stress intensity factor prediction model based on machine learning;

[0066] S4. Adjust the hyperparameters of the neural network and verify the fitting effect of the stress intensity factor prediction model based on machine learning according to the evaluation index;

[0067] S5. Based on the stress intensity factor prediction model based on machine learning, establish a visual interactive interface for input of crack parameters, input of load parameters, input of neural network hyperparameters, output of the image of the load-bearing structure with a surface crack, output of the prediction result, and output of the comparison result of the empirical formula.

[0068] First, a finite element simulation model with a surface crack needs to be established, and the stress intensity factor at the front of the surface crack is solved. Developing a stress intensity factor prediction model based on machine learning requires a dataset, which is used to train the machine learning model. The quality and quantity of the data determine the achievable accuracy. Therefore, it is very important to construct a good dataset, which includes a combination of input data and target data. Then, based on the generated machine learning dataset, a backpropagation neural network (BP) model is built, and a hybrid optimization of the particle swarm optimization algorithm (PSO) and genetic algorithm (GA) is adopted to form a PSOGA-BP prediction model.

[0069] Develop a graphical user interface (GUI) to facilitate access to and visualization of relevant problems and existing machine learning (ML) solutions; design the PSOGA-BP prediction model into a graphical user interface using MATLAB. This graphical user interface can realize the input of crack parameters, the setting of hyperparameters of the PSOGA-BP neural network training model, the output of the stress intensity factor of the surface crack, the schematic diagram of a cylinder with a surface semi-elliptical crack, and the comparison between the stress intensity factor calculation results of the Newman-Raju empirical formula and the prediction results. In the graphical user interface, users do not need to learn complex codes but can operate through the graphical objects therein, greatly reducing the usage threshold.

[0070] Specifically, in this embodiment, the load-bearing structure includes a cylindrical structure, and also includes load-bearing structures of different shapes in other actual usage scenarios, such as cube, cuboid, cone, triangular prism and other structures; the load-bearing structure in this embodiment and the following text is a cylindrical structure;

[0071] The positions of the surface semi-elliptical cracks established on the cylinder include the axial direction of the outer surface, the axial direction of the inner surface, the circumferential direction of the outer surface, and the circumferential direction of the inner surface. The parameter range includes circular cracks (a / c = 1) to long elliptical cracks (a / c = 0.1), shallow cracks (9a / t = 0.1) to deep cracks (a / t = 0.9), thin-walled cylinders (t / r = 0.1) to thick-walled cylinders (t / r = 0.9). The established finite element models include the cylinder models with surface cracks under internal pressure load and uniform tensile load at both ends, with a total of 3600 different samples.

[0072] In some embodiments, establishing a finite element simulation model with surface cracks includes the following steps:

[0073] S11, establish a finite element model of the load-bearing structure through ABAQUS software, and export the output file of the finite element model of the load-bearing structure;

[0074] S12, insert a semi-elliptical surface crack into the model of the output file of the load-bearing structure model through FRANC3D software, and output the finite element model of the load-bearing structure with surface cracks.

[0075] Specifically, due to the large number of samples and the wide range of dimensions included, the finite element software ANAQUS and the crack analysis software FRANC3D are used for joint simulation. First, after completing the finite element model of the cylinder in ABAQUS, the output file (.inp) is generated and imported into FRANC3D.

[0076] In some embodiments, inserting a semi-elliptical surface crack into the model of the output file of the cylinder structure model through FRANC3D software includes the following steps:

[0077] S121, divide the finite element model of the load-bearing structure into an overall model and a local model;

[0078] S122, define the short semi-axis size and the long semi-axis size of the semi-elliptical surface crack, and introduce the semi-elliptical surface crack into the local model to form a local model with surface cracks;

[0079] S123, mesh the local model with surface cracks;

[0080] S124, merge the overall model and the local model with surface cracks to form a finite element model of the load-bearing structure with surface cracks.

[0081] Specifically, the complete finite element model is divided into two parts: the global model and the local model. Then, a semi-elliptical surface crack is selected, and the depth a and the major semi-axis dimension c of the initial semi-elliptical surface crack are defined. The semi-elliptical surface crack is introduced into the local model. Then, the local model with the crack is meshed by FRANC3D. Finally, the global model and the local model are combined to form a complete cylindrical finite element model containing the initial crack, as shown in Figure 5 the attached Figure 5 . (a) in the attachment is the cylindrical finite element model with a circumferential semi-elliptical crack on the outer surface, and (b) in the attachment is the cylindrical finite element model with an axial semi-elliptical crack on the outer surface. Figure 5

[0082] Stress intensity factor formula for an axial semi-elliptical crack on the outer (inner) surface of the cylinder:

[0083]

[0084] P refers to the internal pressure load, R i refers to the inner radius of the cylinder, a refers to the minor semi-axis of the crack, c refers to the major semi-axis of the crack, t refers to the thickness of the cylinder, refers to the parametric angle of the ellipse, Q refers to the second kind of elliptic integral, F e(i) refers to the surface crack boundary correction factor.

[0085] Stress intensity factor formula for a circumferential semi-elliptical crack on the outer (inner) surface of the cylinder:

[0086]

[0087] σ is the axial tensile load.

[0088] The finite element simulation model is jointly established by the finite element software ABAQUS and the crack analysis software FRANC3D. ABAQUS is responsible for establishing the structure that bears the load, while FRANC3D focuses on inserting the crack. The joint simulation not only improves the calculation accuracy but also improves the modeling efficiency.

[0089] In some embodiments, the following steps are included to generate a finite element simulation dataset by performing dimensionless processing on the size of the surface crack, the size of the structure that bears the load, and the stress intensity factor at the front edge of the surface crack in the finite element simulation dataset:

[0090] S21. Nondimensionalize the short semi-axis size of the crack, the long semi-axis size of the crack, the thickness of the load-bearing structure, the inner radius of the load-bearing structure, and the stress intensity factor. Use the normalized distance of the crack front to replace the parametric angle of the ellipse to form nondimensional quantities. The nondimensional quantities are specifically: the ratio of the short semi-axis of the crack to the long semi-axis of the crack, the ratio of the short semi-axis of the crack to the thickness of the load-bearing structure, the ratio of the thickness of the load-bearing structure to the characteristic size of the load-bearing structure, the normalized distance of the crack front, and the surface crack boundary correction factor. In this embodiment, the ratio of the thickness of the load-bearing structure to the characteristic size of the load-bearing structure is: the ratio of the thickness of the load-bearing structure to the inner radius size of the cylinder.

[0091] S22. Generate an axial semi-elliptical crack data set and a circumferential semi-elliptical crack data set.

[0092] Specifically, to more conveniently establish a machine learning data set, the angle is replaced by the normalized distance of the crack front . The five independent nondimensional quantities are all related. Therefore, the axial semi-elliptical crack data set adopts the following structure:

[0093] Input variables:

[0094]

[0095] Output target:

[0096]

[0097] The circumferential semi-elliptical crack data set adopts the following structure:

[0098] Input variables:

[0099]

[0100] Output target:

[0101]

[0102] The machine learning data set is generated by nondimensionalizing the crack geometric features and the stress intensity factor. A finite element model is established by combining ABAQUS and FRANC3D to calculate the stress intensity factor, and it is compared with the results of the Newman-Raju empirical formula. An error of less than 5% between the two indicates that the finite element model is reliable, ensuring the modeling accuracy and data reliability.

[0103] A finite element model is established through the combined simulation of ABAQUS and FRANC3D. ABAQUS is responsible for the modeling of the load-bearing structure and the setting of boundary conditions, while FRANC3D focuses on the insertion of cracks, mesh generation, and the accurate solution of stress intensity factors. This method avoids the problems of insufficient accuracy of empirical formulas, limitations of specific crack structures, and the time-consuming complex crack mesh generation when using ABAQUS alone. The combined simulation not only improves the calculation efficiency and saves time costs but also can quickly generate a large amount of high-precision model data, providing reliable support for the training of subsequent machine learning models.

[0104] In some embodiments, optimizing the backpropagation neural network through a hybrid optimization algorithm includes the following steps:

[0105] S31, Optimize the weights and biases of the backpropagation neural network through the particle swarm algorithm to obtain the optimal combination of weights and biases of the backpropagation neural network;

[0106] S32, Apply the hybridization and mutation operations of the genetic algorithm to the particle swarm algorithm to enhance the population diversity and global search ability.

[0107] In some embodiments, step S1 is:

[0108] S11a, Based on the position of the surface crack in the load-bearing structure, generate a finite element simulation model containing surface cracks at different positions;

[0109] Among them, the position of the surface crack in the load-bearing structure is specifically the outer surface axis of the load-bearing structure, the inner surface axis of the load-bearing structure, the outer surface circumferential of the load-bearing structure, and the inner surface circumferential of the load-bearing structure;

[0110] Among them, the parameter range includes circular cracks to long elliptical cracks, shallow cracks to deep cracks, thin-walled load-bearing structures to thick-walled load-bearing structures;

[0111] Please refer to the appendix Figure 4 , the appendix Figure 4 is a schematic diagram of the load-bearing structure with a surface semi-elliptical crack disclosed in the embodiments of the present application. Specifically, in this embodiment, the load-bearing structure is a cylindrical structure, and the appendix Figure 4 (a) is a schematic diagram of a cylinder with a circumferential semi-elliptical crack on the outer surface, and the appendix Figure 4 (a) is a schematic diagram of a cylinder with a circumferential semi-elliptical crack on the inner surface, and the appendix Figure 4 (a) is a schematic diagram of a cylinder with an axial semi-elliptical crack on the outer surface, and the appendix Figure 4 (a) is a schematic diagram of a cylinder with an axial semi-elliptical crack on the inner surface; in the figure a is the short semi-axis of the crack, c is the long semi-axis of the crack, t is the thickness of the cylinder, and R i is the inner radius of the cylinder, and Ro is the outer radius of the cylinder.

[0112] The machine learning data set established in the present invention is generated by the finite element model jointly established by ABAQUS and FRANC3D, ensuring the accuracy and reliability of the stress intensity factor results. The dimensionless processing of crack geometric features and stress intensity factors covers a wide range of crack sizes and positions, improving the representativeness and applicability of the data set.

[0113] Specifically, the machine learning model is a PSOGA-BP neural network established by MATLAB. The BP neural network mainly consists of an input layer, a hidden layer, and an output layer. The present invention attempts to construct BP neural networks with different numbers of hidden layers and neurons, and evaluate the model performance according to the adjusted results to ensure the accuracy and generalization ability of the model. Aiming at the problems of slow speed, low accuracy, and easy to fall into local optimum in the training process of the traditional BP neural network, the present invention combines the particle swarm optimization algorithm (PSO) and the genetic algorithm (GA) for optimization. The PSO and GA are used to optimize the BP neural network. Since the PSO algorithm has a fast convergence speed, the PSO algorithm is used to optimize the weights and biases of the BP neural network to help find the optimal combination of weights and biases, improving the training speed of the BP neural network. The crossover and mutation operations of GA are applied to the PSO algorithm to enhance the population diversity and global search ability, improving the convergence speed and prediction accuracy of the BP neural network.

[0114] After establishing the neural network, adjusting the hyperparameters is essential. This process can train a more efficient machine learning model more scientifically. Among them, the learning rate and the number of neurons in the hidden layer will have a greater impact on the fitting accuracy. Using the exhaustive method to adjust the hyperparameters of the neural network, it is found that when the learning rate is 0.00001 and there are three hidden layers (15-10-10), there is a higher fitting accuracy, as shown in Figure 6 shown in Figure 6 is a schematic diagram of the regression value between the predicted value and the simulated value of the PSOGA-BP neural network model.

[0115] Finally, through the graphical user interface of MATLAB, users can directly complete key functions through graphical operations without writing or learning complex codes, greatly reducing the usage threshold and learning cost. The functions include intuitively displaying the predicted value of the stress intensity factor to help users understand the stress intensity factor at the crack front. The GUI provides evaluation indexes such as root mean square error (RMSE), correlation coefficient (R 2 ) etc., so as to intuitively judge the prediction effect of the model, and allow users to adjust hyperparameters such as learning rate and hidden layer structure through the interface to optimize the performance of the neural network. The geometric shape and position of the surface crack of the structure under load are displayed graphically in the GUI, intuitively assisting users to understand the crack distribution, such asFigure 5 as shown

[0116] Specifically, please refer to the appendix Figure 7 , the appendix Figure 7 is the schematic diagram of the graphical user interface in this embodiment. In the appendix Figure 7 , at position 1 is to display the prediction result of the stress intensity factor at the crack front, in the appendix Figure 7 , at position 2 is to display the performance evaluation of the prediction index model, in the appendix Figure 7 , at position 3 is the setting of the hyperparameters of the neural network, in the appendix Figure 7 , at position 4 is the prediction of the stress intensity factor of the axial semi-elliptical crack on the surface of the cylinder, in the appendix Figure 7 , at position 5 is the prediction of the stress intensity factor of the circumferential semi-elliptical crack on the surface of the cylinder, in the appendix Figure 7 , at position 6 is to display the image of the cylinder with a semi-elliptical crack, in the appendix Figure 7 , at position 7 is to solve the stress intensity factor of the internal surface axial semi-elliptical crack by the empirical formula Newman-Raju, in the appendix Figure 7 , at position 8 is to solve the stress intensity factor of the external surface axial semi-elliptical crack by the empirical formula Newman-Raju. When training a new model Figure 7 , at position 9 is to solve the stress intensity factor of the external surface circumferential semi-elliptical crack by the empirical formula Newman-Raju, in the appendix Figure 7 , at position 10 is to calculate the stress intensity factor of the external surface circumferential semi-elliptical crack;

[0117] In some embodiments, the evaluation criteria in step S4 include at least one of the coefficient of determination (R 2 ), mean absolute error (MAE), mean bias error (MBEC), and mean absolute percentage error (MAPE). In this embodiment, the evaluation criterion is the correlation coefficient (R 2 ).

[0118] The user needs to set the hyperparameters in the graphical user interface (as shown at position 3 in the appendix Figure 7 ), including the number of neurons in the hidden layer, the number of training times, the learning rate, the convergence error, the proportion of the training set, the activation function, the number of population iterations, the population size, the particle velocity boundary and position boundary, the crossover probability, and the mutation probability. After the setting is completed, import the training data (as shown at position 2 in the appendix Figure 7 ), start training. After the training is completed, the root mean square error (RMSE) and the correlation coefficient (R 2) and other evaluation metrics to observe the performance of the model. Save the neural network model whose fitting accuracy meets the requirements for convenient direct calling. The prediction models for the stress intensity factor of axial semi-elliptical cracks on the outer surface of the cylinder and the inner surface, net1.mat and net2.mat, and the prediction models for the stress intensity factor of circumferential semi-elliptical cracks on the outer surface of the cylinder and the inner surface, net3.mat and net4.mat. Just input the crack size and load conditions (as shown at 4 places in the appendix Figure 7 ), and the prediction of the stress intensity factor of surface cracks can be immediately realized.

[0119] The graphical user interface (GUI) is the design implementation of encapsulating and visualizing the PSOGA-BP prediction model constructed in the MATLAB environment. Through this interface, users can directly complete functions such as model parameter setting, data input, training and running, and result visualization through graphical operations without in-depth programming. This GUI integrates the core algorithms of the particle swarm optimization algorithm (PSO), genetic algorithm (GA), and backpropagation BP neural network, provides intuitive parameter adjustment options and real-time feedback functions, and improves the convenience and efficiency of model use. The GUI developed based on MATLAB provides a visual interactive design, enabling users to complete machine learning model parameter setting, data input, training execution, and result display without in-depth programming, greatly reducing the usage threshold and improving the operation efficiency.

[0120] Example 2

[0121] A surface crack stress intensity factor prediction device based on machine learning, the device includes:

[0122] A dataset acquisition module 100, which is used to receive the finite element simulation model with surface cracks established by the user, calculate the stress intensity factor at the front edge of the surface crack, verify the stress intensity factor of the finite element simulation model through the Newman-Raju empirical formula, and generate a finite element simulation dataset;

[0123] A data preprocessing module 200, which is used to perform dimensionless processing on the size of the surface crack, the size of the load-bearing structure, and the stress intensity factor at the front edge of the surface crack in the finite element simulation dataset to generate a machine learning dataset;

[0124] A machine learning model generation module 300, which is used to optimize the backpropagation neural network through a hybrid optimization algorithm to construct a stress intensity factor prediction model based on machine learning;

[0125] A model training and verification module 400, which is used to adjust the hyperparameters of the neural network and verify the fitting effect predicted by the stress intensity factor prediction model based on machine learning according to the evaluation metrics;

[0126] The graphical user interface output module 500 is used to establish a visual interaction interface for input of crack parameters, input of load parameters, input of neural network hyperparameters, output of an image of a load-bearing structure with a surface crack, output of prediction results, and output of comparison results of empirical formulas based on a machine learning-based stress intensity factor prediction model.

[0127] In some embodiments, establishing a finite element simulation model with a surface crack includes:

[0128] The load-bearing structure modeling module 110 is used to establish a finite element model of the load-bearing structure through ABAQUS software and export the output file of the finite element model of the load-bearing structure;

[0129] The crack insertion modeling module 120 is used to insert a semi-elliptical surface crack into the model of the output file of the load-bearing structure model through FRANC3D software and output a finite element model of the load-bearing structure with a surface crack.

[0130] Embodiment III

[0131] An electronic device includes a memory and a processor, wherein:

[0132] The memory is used to store a computer program;

[0133] The processor is used to execute the computer program to implement the machine learning-based surface crack stress intensity factor prediction method disclosed in the foregoing embodiments.

[0134] For the specific process of the above-mentioned machine learning-based surface crack stress intensity factor prediction method, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details will not be elaborated herein.

[0135] Moreover, as a carrier for storing resources, the memory can be a read-only memory, a random access memory, a disk, or an optical disc, etc., and the storage method can be temporary storage or permanent storage.

[0136] Embodiment IV

[0137] A computer-readable storage medium stores a computer program, and the computer program is used to be run by a processor to implement the machine learning-based surface crack stress intensity factor prediction method disclosed in the foregoing embodiments.

[0138] For the specific process of the above-mentioned machine learning-based surface crack stress intensity factor prediction method, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details will not be elaborated herein.

[0139] Embodiment V

[0140] A computer program product includes a computer program / instructions, and when the computer program / instructions are executed by a processor, they implement the machine learning-based surface crack stress intensity factor prediction method disclosed in the foregoing embodiments.

[0141] For the specific process of the above-mentioned machine learning-based surface crack stress intensity factor prediction method, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details will not be repeated here.

[0142] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and reference can be made to the method part for the relevant parts.

[0143] The steps of the methods or algorithms described in connection with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0144] The above has introduced in detail a machine learning-based surface crack stress intensity factor prediction method, device, product, equipment, and medium provided by this application. Specific examples are used herein to elaborate on the principle and implementation manner of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for predicting surface crack stress intensity factor based on machine learning, characterized in that: The following steps are involved: S1, establish a finite element simulation model with surface cracks, calculate the stress intensity factor of the surface crack front, verify the stress intensity factor of the finite element simulation model through the Newman-Raju empirical formula, and generate a finite element simulation data set; S2, dimensionlessly processing the size of the surface crack, the size of the load-bearing structure, and the stress intensity factor of the surface crack front in the finite element simulation data set to generate a machine learning data set; S3, optimizes the back propagation neural network through a hybrid optimization algorithm and constructs a stress intensity factor prediction model based on machine learning; S4, adjust the neural network hyperparameters and verify the fitting effect of the stress intensity factor prediction model based on machine learning according to the evaluation indicators; S5, a stress intensity factor prediction model based on machine learning, establishes a visual interactive interface for crack parameter input, load parameter input, neural network hyperparameter input, image output of the load-bearing structure with surface cracks, prediction result output, and empirical formula comparison result output.

2. The method for predicting surface crack stress intensity factor based on machine learning as claimed in claim 1, characterized in that: The establishment of the finite element simulation model containing surface cracks comprises the following steps: S11, establishing a finite element model of the load-bearing structure through ABAQUS software, and exporting an output file of the finite element model of the load-bearing structure; S12, inserting a semi-elliptical surface crack into the model of the load-bearing structure model output file through FRANC3D software, and outputting a finite element model of the load-bearing structure containing the surface crack.

3. The method for predicting surface crack stress intensity factor based on machine learning as claimed in claim 2, characterized in that: The method of inserting a semi-elliptical surface crack into the model of the load-bearing structure model output file by FRANC3D software comprises the following steps: S121, dividing the finite element model of the load-bearing structure into an overall model and a local model; S122, defining the short semi-axis size and the long semi-axis size of the semi-elliptical surface crack, introducing the semi-elliptical surface crack into the local model, and forming a local model containing the surface crack; S123, meshing the local model with surface cracks; S124, merging the overall model and the local model containing the surface cracks to form a finite element model of the load-bearing structure containing the surface cracks.

4. The method for predicting surface crack stress intensity factor based on machine learning as claimed in claim 1, characterized in that: The dimensionless processing of the size of the surface crack, the size of the load-bearing structure and the stress intensity factor of the surface crack front in the finite element simulation data set to generate the finite element simulation data set comprises the following steps: S21, performing dimensionless processing on the crack semi-minor axis size, the crack semi-major axis size, the load-bearing structure thickness, the load-bearing structure inner radius, and the stress intensity factor, replacing the parameter angle of the ellipse with the normalized distance of the crack front edge, and forming a dimensionless quantity, wherein the dimensionless quantity is specifically: the ratio of the crack semi-minor axis to the crack semi-major axis size, the ratio of the crack semi-minor axis to the load-bearing structure thickness, the ratio of the load-bearing structure thickness to the load-bearing structure characteristic size, the crack front edge normalized distance, and the surface crack boundary correction factor; S22, generate axial semi-elliptical crack data set and circumferential semi-elliptical crack data set.

5. The method for predicting surface crack stress intensity factor based on machine learning as claimed in claim 1, characterized in that: The optimization of the back propagation neural network by the hybrid optimization algorithm comprises the following steps: S31, optimizing the weights and biases of the back-propagation neural network through the particle swarm algorithm, and obtaining the optimal combination of weights and biases of the back-propagation neural network; S32, by applying the hybridization and mutation operations of genetic algorithms to particle swarm optimization, the population diversity and global search capabilities are enhanced.

6. The method for predicting surface crack stress intensity factor based on machine learning as claimed in claim 1, characterized in that: The step S1 is: S11a, generating a finite element simulation model containing surface cracks at different positions based on the position of the surface cracks in the load-bearing structure; The position of the surface crack in the load-bearing structure is specifically the axial direction of the outer surface of the load-bearing structure, the axial direction of the inner surface of the load-bearing structure, the circumferential direction of the outer surface of the load-bearing structure, and the circumferential direction of the inner surface of the load-bearing structure; Among them, the parameter range includes circular cracks to long elliptical cracks, shallow cracks to deep cracks, and thin-walled load-bearing structures to thick-walled load-bearing structures; S12a, the finite element simulation model with surface cracks includes a load-bearing structure model with surface cracks under the action of internal pressure load and uniform tensile load at both ends.

7. The method for predicting surface crack stress intensity factor based on machine learning as claimed in claim 1, characterized in that: The evaluation criteria in step S4 include at least one of a coefficient of determination, a mean absolute error, a mean deviation error, and a mean absolute percentage error.

8. A device for predicting surface crack stress intensity factor based on machine learning, characterized in that: The device comprises: The data set acquisition module is used to receive the finite element simulation model with surface cracks established by the user, calculate the stress intensity factor of the surface crack front, verify the stress intensity factor of the finite element simulation model through the Newman-Raju empirical formula, and generate a finite element simulation data set; A data preprocessing module is used to perform dimensionless processing on the size of surface cracks, the size of the load-bearing structure, and the stress intensity factor of the surface crack front in the finite element simulation data set to generate a machine learning data set; A machine learning model generation module is used to optimize the back propagation neural network through a hybrid optimization algorithm and build a stress intensity factor prediction model based on machine learning; Model training and validation module, used to adjust the neural network hyperparameters and verify the fitting effect of the stress intensity factor prediction model based on machine learning according to the evaluation indicators; The graphical user interface output module is used for the stress intensity factor prediction model based on machine learning, and establishes a visual interactive interface for crack parameter input, load parameter input, neural network hyperparameter input, image output of the load-bearing structure with surface cracks, prediction result output, and empirical formula comparison result output.

9. The device for predicting surface crack stress intensity factor based on machine learning as claimed in claim 8, characterized in that: The establishment of a finite element simulation model containing surface cracks comprises: The load-bearing structure modeling module is used to establish a finite element model of the load-bearing structure through ABAQUS software and export the finite element model output file of the load-bearing structure; The crack insertion modeling module is used to insert a semi-elliptical surface crack into the model of the load-bearing structure model output file through FRANC3D software, and output a finite element model of the load-bearing structure containing the surface crack.

10. An electronic device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store the computer program; The processor is used to execute the computer program to implement the surface crack stress intensity factor prediction method based on machine learning as described in any one of claims 1 to 7.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program is used to be executed by a processor to implement the surface crack stress intensity factor prediction method based on machine learning as described in any one of claims 1 to 7.

12. A computer program product, characterized in that It includes a computer program / instruction, which, when executed by a processor, implements the surface crack stress intensity factor prediction method based on machine learning as described in any one of claims 1 to 7.

Images