Welding structure fatigue life prediction method and system based on dual-drive strategy

Abstract

The invention discloses a welding structure fatigue life prediction method and system based on a dual-drive strategy, and belongs to the field of engineering welding structure fatigue performance analysis, and the method comprises the steps: generating a target function model based on a deep convolutional neural network DCNN through introducing physical constraints, and obtaining a fatigue life prediction model; the physical constraint is used for realizing physical constraints of material strength, average stress, geometric shape of a welding seam, load conditions and a size effect on a target function level; according to the fatigue performance influence factors of the engineering welding structure, a fatigue performance data set is obtained, model training is conducted through the target function model, and a fatigue life prediction model is constructed and used for predicting the fatigue life of the engineering welding structure. According to the method, a large amount of manpower and material resources consumed by the fatigue test are greatly reduced, a new thought and method are provided for fatigue performance analysis of major engineering equipment and structures, and powerful support is provided for fatigue design of engineering welding structures.

Description

Technical Field

[0001] The present invention relates to the technical field of fatigue performance analysis of engineering welded structures, and in particular to a method and system for predicting fatigue life of welded structures based on a dual-drive strategy. Background Art

[0002] At present, there are higher requirements for the reliability of engineering equipment and structures. Given that fatigue fracture is one of the main failure modes of engineering structures, the study of fatigue performance is becoming increasingly important. With the rapid development of global social economy and industrial technology, welding technology has become an indispensable key technology for engineering equipment and structure manufacturing. However, the fatigue performance of engineering welded structures is difficult to be equivalent to that of the parent material due to stress concentration caused by different geometric shapes and sizes, which has a significant impact on the performance of engineering equipment and structures, making the reliability of welded structures a key link in industrial production and engineering structure manufacturing.

[0003] In the fatigue life prediction of welded structures, changes in multiple conditions have a significant impact on fatigue behavior, resulting in a large fluctuation range of fatigue life in the data set. This series of effects is the main obstacle to building a unified model for fatigue life prediction. Physically driven prediction models and purely data-driven prediction models have difficulty understanding fatigue behavior under multiple conditions with a large span, and data-driven fatigue life prediction methods can improve the above problems.

[0004] However, the existing intelligent prediction methods usually have the limitation of poor generalization ability, which leads to poor results in actual engineering applications. Therefore, it is urgent to design an intelligent prediction technology that can be enhanced by a physical-data dual-driven strategy to better combine the application scenarios of different engineering welding structures, so as to achieve more reliable fatigue life prediction, which brings a new approach to the reliability research of major engineering structures and equipment. Summary of the invention

[0005] In order to solve the above problems, the purpose of the present invention is to provide a welding structure fatigue life prediction technology based on a dual-drive strategy, aiming to achieve accurate and stable fatigue life prediction of engineering welded structural parts under a variety of different service conditions.

[0006] In order to achieve the above technical objectives, the present application provides a method for predicting fatigue life of a welded structure based on a dual-drive strategy, comprising the following steps:

[0007] Based on the deep convolutional neural network DCNN, the objective function model is generated by introducing physical constraints to realize the physical constraints of material strength, average stress, weld geometry, load conditions and size effect at the objective function level;

[0008] According to the factors affecting the fatigue performance of engineering welded structures, the fatigue performance data set is obtained, the model is trained through the objective function model, and a fatigue life prediction model is constructed to predict the fatigue life of engineering welded structures.

[0009] Preferably, in the process of constructing the target model, the target model is expressed as:

[0010]

[0011] In the formula, K t represents the stress concentration factor, E represents the elastic modulus, σ a represents the stress amplitude, Δσ represents the stress range, H c and h c They represent the cyclic hardening coefficient and cyclic hardening exponent respectively.

[0012] Preferably, when obtaining the cyclic hardening index h c During the process, the cyclic hardening index h c It is expressed as:

[0013]

[0014] Where b and c represent the fatigue strength index and fatigue ductility index, respectively.

[0015] Preferably, in the process of obtaining, it is expressed as:

[0016]

[0017] In the formula, σ′ f , ε′ f represent fatigue strength coefficient and fatigue ductility coefficient respectively.

[0018] Preferably, in the process of constructing a fatigue performance data set, an analysis is performed based on the service status of the engineering welded structure to obtain the factors affecting fatigue performance, construct an original fatigue performance data set, and after data enhancement through the borderline synthetic minority class oversampling technology Borderline-SMOTE, a fatigue performance data set is generated.

[0019] Preferably, in the process of constructing the original fatigue performance data set, based on the data set corresponding to the fatigue performance influencing factors, the original fatigue performance data set is generated after data processing through data cleaning, data preprocessing and data screening, wherein the data whose deviation between the fatigue life value and the value on the SN curve exceeds 5 times the average deviation of all data in the data set, and the data whose deviation between the fatigue life value and the life value on the SN curve exceeds 10 times are removed.

[0020] Preferably, in the process of model training, based on the fatigue performance data set, the importance of each physical and chemical influencing factor is analyzed through the extreme gradient boosting algorithm, sensitivity analysis method, Pearson correlation coefficient and maximum information coefficient, and the importance analysis result is used as the initial weight of the model for model training.

[0021] Preferably, in the process of constructing the fatigue performance data set, if the data volume is sufficient, the Borderline-SMOTE method is used to implement data enhancement; if the data volume is insufficient, the generative adversarial network or genetic algorithm method is used to increase the data volume.

[0022] Preferably, in the process of obtaining the factors affecting fatigue performance, based on different load forms, the load spectrum is converted into nominal stress to obtain the factors affecting fatigue performance.

[0023] The present invention discloses a welding structure fatigue life prediction system based on a dual-drive strategy, the system is used to implement a welding structure fatigue life prediction method, the system comprises:

[0024] The model building module, based on the deep convolutional neural network DCNN, generates the objective function model by introducing physical constraints, which is used to realize the physical constraints of material strength, average stress, weld geometry, load conditions and size effect at the objective function level;

[0025] A data acquisition module is used to obtain fatigue performance data sets based on the fatigue performance influencing factors of engineering welded structures;

[0026] The prediction module is used to train the model based on the objective function model through the fatigue performance data set, build a fatigue life prediction model, and predict the fatigue life of engineering welded structures.

[0027] The present invention discloses the following technical effects:

[0028] The present invention can capture and analyze important fatigue life influencing factors of engineering welded structures under different service conditions, and can comprehensively consider the effects of different influencing factors on fatigue life, thereby achieving fatigue life prediction with better adaptability.

[0029] The present invention has good performance in direct prediction and indirect prediction of fatigue life value and prediction of SN curve, and has good applicability in actual engineering welded structures; thereby greatly reducing the large amount of manpower and material resources consumed by fatigue tests, providing new ideas and methods for fatigue performance analysis of major engineering equipment and structures, and providing strong support for fatigue design of engineering welded structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 is a flow chart of a method provided by an embodiment of the present invention;

[0032] Figure 2 is a prediction result diagram of a welded hull structure provided by Example 1 of the present invention;

[0033] Figure 3 is a diagram showing the prediction results of the support column structure of an offshore cruise ship provided by Example 2 of the present invention;

[0034] Figure 4 is a prediction result diagram of a marine steel bridge welding structure provided by Example 3 of the present invention;

[0035] Figure 5 This is a diagram showing the prediction results of the electric locomotive bogie structure provided by Example 4 of the present invention. DETAILED DESCRIPTION

[0036] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0037] like Figure 1-5 As shown, the present invention provides an intelligent prediction and analysis method for fatigue life of engineering welded structures based on a physical-data dual-driven strategy. Figure 1 As shown, including the following:

[0038] 1. Analyze the service status of the structure to be tested and collect the factors that can be considered to affect the fatigue performance (quantitative or qualitative, numerical or non-numerical features). After the collected influencing factors are determined, they can be used as input features of the physical-data dual-driven prediction model.

[0039] 2. According to the collected fatigue performance influencing factors, based on the establishment of fatigue performance data set and a series of preprocessing methods, a data set suitable for the welded structure to be tested is built. And the data set is enhanced based on the borderline synthetic minority class oversampling technology (Borderline-SMOTE).

[0040] When constructing the original fatigue performance data set, some data are obtained by summarizing the literature. In the process, there will be differences in test conditions and test methods of different studies, incomplete information such as test conditions and parameter settings, missing data, and diverse data formats. The above problems will affect the consistency and quality of the data set, and then affect the performance of the fatigue life prediction method. Therefore, it is necessary to perform data cleaning, data preprocessing, and data screening on the established original data set to improve the accuracy and reliability of the fatigue life prediction model. Among them, the goal of the cleaning, preprocessing, and screening process of the original data set is to identify and correct inconsistencies and errors in the data set, fill in missing values, unify data formats, and eliminate data that does not conform to the objective laws of life distribution, rather than creating an idealized data set. Based on the above process, it can be ensured that the processed fatigue performance data set has the advantages of sufficient data volume, rich features, and high quality, laying a data foundation for building a reliable fatigue life prediction model.

[0041] Fatigue data that does not conform to objective laws will cause the prediction model to over-adapt to these abnormal data characteristics and ignore the actual fatigue behavior pattern, thereby increasing the risk of overfitting and causing inaccurate fatigue life prediction results. Removing data that does not conform to objective laws can make the model training and prediction process more in line with actual fatigue behavior and simplify the training process, thereby enhancing the generalization ability of the prediction model on new data. In the process of removing data that does not conform to objective laws, formulas (1) and (2) are proposed as criteria to analyze the data in the original data set, and fatigue data that does not meet any of the following two formulas is removed. Among them, formula (1) is used to remove data whose deviation between the fatigue life value to be determined and the value on the SN curve exceeds 5 times the average deviation of all data in the data set, and formula (2) is used to remove data whose deviation between the fatigue life value to be determined and the life value on the SN curve exceeds 10 times:

[0042]

[0043] Among them, N point,c is the fatigue life value to be determined, N S-N,c is the fatigue life value in the SN curve when the stress range is the same as the data to be determined, n a is the number of all fatigue performance data in the original data set, N point,i and N S-N,iThey respectively represent the current fatigue life value and the fatigue life value in the SN curve with the same stress range as the current data.

[0044] 3. The fatigue life prediction model of engineering welded structures based on physical-data dual drive takes the deep convolutional neural network (DCNN) as the basic framework. On this basis, the physical constraint scheme is first determined to implement physical feature engineering on the established fatigue performance data set. In the current study, the physical-driven Morrow, SWT and IG models can consider the influence of average stress on fatigue life, thereby realizing the fatigue life prediction of welded joints. Therefore, the above physical model is used to constrain the data-driven method, and its calculation process can be expressed by equations (3), (4) and (5):

[0045]

[0046]

[0047] Among them, σ m is the mean stress, E is the elastic modulus, σ a is the stress amplitude, σ′ f , ε′ f , b and c are fatigue strength coefficient, fatigue ductility coefficient, fatigue strength index and fatigue ductility index respectively, H c and h c They are the cyclic hardening coefficient and cyclic hardening exponent respectively.

[0048] Furthermore, by taking equation (6) as the objective function of the data-driven model, the fusion of the introduced physical model and the data-driven method is realized, and the physical constraints of material strength, average stress, weld geometry, load conditions and size effect at the objective function level are realized. Through the fusion of the two, the fatigue life prediction model constructed can learn the fatigue behavior of welded joints under complex working conditions from the data set and ensure that the prediction results follow objective physical laws. In addition, the process control of the data-driven method under different fatigue conditions is realized through the physicalization of functions at different levels, which enhances the operability of the fatigue life prediction model.

[0049]

[0050] In formula (6), K t represents the stress concentration factor, and Δσ represents the stress range.

[0051] 4. Based on the constructed fatigue performance dataset and influencing factor analysis methods such as the eXtreme Gradient Boosting (XGBoost) algorithm, sensitivity analysis method, Pearson Correlation Coefficient (PCC) and Maximum Information Coefficient (MIC), the importance of each physical and chemical influencing factor is analyzed to provide data support for the manufacturing process optimization of the structure to be tested and the determination of the life extension method. The importance analysis results are used as the initial weight of the DCNN model to build a fatigue life prediction model for engineering welded structures based on physical and data dual drive.

[0052] 5. Take the service characteristics of the structure to be tested as input, and select direct or indirect fatigue life prediction, SN curve prediction or a combination of schemes according to actual needs to realize fatigue life prediction of the actual welded structure. Among them, the inputs of direct or indirect fatigue life prediction and SN curve prediction are all factors affecting fatigue life, the output of the direct fatigue life prediction method is fatigue life, and the output of the indirect fatigue life prediction and SN curve prediction method is the parameters C and m of the SN curve.

[0053] In addition, since the actual working conditions of engineering welded structures are quite complex, in the above fatigue life prediction process, the prediction architecture under different working conditions needs to be properly optimized to better adapt to multi-scale, multi-environment, and multi-state working conditions. Figure 1 The numbers marked in the figure are the issues that were considered in the process of building a prediction architecture based on physical and data dual drive. The following considers the relevant issues and provides corresponding solutions:

[0054] ① In the process of determining the input parameters, due to the different working conditions of the structure to be tested, the fatigue performance influencing factors that can be collected are not completely consistent. Therefore, the influencing factors shown in the figure are mainly used for reference and do not need to be completely consistent. Specifically, when the number of influencing factors collected in the data set is greater than 5, the constructed prediction architecture can show better prediction performance.

[0055] ② Since the matching forms of different structures are different, and the structure to be tested usually contains a combination of multiple butt joints, T-type joints and cross joints, the parameters used to represent the influence of plate thickness can be expressed by a vector group. A feasible solution is: according to the welding sequence of the structure to be tested, the thickness of the main plate and rib plate of each joint can be fully considered. The parameters representing the influence of plate thickness can be expressed by the vector group shown in equations (7) and (8):

[0056]

[0057] vT1 =[T1,T2,...,T n ] (8)

[0058] Among them, v T1 is the vector representing the thickness characteristics of the nth weld, T n Indicates the thickness of the main plate / rib of the weld.

[0059] ③ The fatigue load forms of different structures to be tested are not the same, but the consistency of the training set and the test data should be maintained. The current prediction architecture should understand the load form as the relationship between numerical characteristics and fatigue life. Therefore, load forms such as wave loads should be converted into nominal stress using load spectra, so as to achieve more reliable fatigue life prediction.

[0060] ④ Data enhancement is very important for improving the quality of fatigue performance datasets. If the data volume of the original fatigue performance dataset is large enough, it is recommended to use the Borderline-SMOTE method to implement data enhancement to ensure its balance; if the data volume of the original fatigue performance dataset is insufficient, it is recommended to use generative adversarial networks or genetic algorithms to increase the data volume of the dataset accordingly.

[0061] ⑤The implementation of physical feature engineering is of great significance for enhancing the physical meaning of the established prediction framework. The implementation of feature engineering can select different physical laws and constraints according to the actual working conditions of the structure to be tested, and process them into different forms of physical influencing factors.

[0062] ⑥ There are various methods for analyzing the importance of physical and chemical influencing factors. Considering the combination effect with the DCNN model, it is recommended to use the XGBoost algorithm combined with cross-validation, PCC and MIC methods to analyze and verify each influencing factor. On this basis, the importance analysis results can provide important support for the optimization of the manufacturing process of the structure to be tested and the determination of the life extension method.

[0063] ⑦ In the process of optimizing the physical activation function and loss function, it is recommended to use the above optimization scheme of the physical function for corresponding settings. If there are additional physical constraints, further optimization can be performed on this basis.

[0064] ⑧The fatigue life prediction scheme of the welded structure can be determined according to the actual service requirements of the structure to be tested. Specifically, the direct fatigue life prediction method can reliably predict the fatigue life of the welded joint under a variety of complex service conditions, and the indirect fatigue life prediction method can reliably predict the SN curve of the welded joint under a variety of complex service conditions and the fatigue life with a smaller fluctuation range. On this basis, according to the actual evaluation needs of the structure to be tested, the fatigue life prediction scheme of the welded structure is determined through direct or indirect prediction of fatigue life, SN curve prediction, or a combination of the above different methods.

[0065] Example 1: Figure 2 As shown, the prediction result of the welded hull structure obtained by the intelligent prediction method for fatigue life of engineering welded structures based on the physical-data dual-drive strategy of the present invention is:

[0066] Erny et al. studied the fatigue performance of hull structures made of DH36 steel welding, and predicted their fatigue life based on the finite element damage model. From this study, the fatigue load information of the hull structure can be obtained as follows: serving in a seawater corrosive environment, the yield strength is 400MPa, the fatigue strength index is 15, the stress ratio is 0.1, the elastic modulus is 210GPa, the Poisson's ratio is 0.3, the fatigue limit is 230MPa, the plate thickness is 8mm, 6.2mm, 5mm, ρ: 1mm, the weld toe radius is 45°, the load type is tension and compression + bending, the ambient temperature is room temperature, and the joint type is a T-joint.

[0067] Based on the above influencing factors, the fatigue life of the key positions of the DH36 steel welded hull structure was predicted. The results showed that the standard deviation of the prediction error (RMSE) of the fatigue life prediction results obtained by the physical-data dual-driven method was 0.08, which was basically consistent with the results obtained by the finite element damage method used in the literature (0.07). In addition, the average error of the fatigue life prediction results obtained based on the established architecture was 0.29, which was 37.0% lower than the average error (0.46) obtained in the literature, indicating that the prediction results obtained by the proposed method are more consistent with the actual results, and the proposed method can accurately predict the fatigue life of welded hull structures in marine corrosive environments.

[0068] Example 2: Figure 3 As shown, the prediction result of the offshore cruise ship support column structure obtained by the intelligent prediction method for fatigue life of engineering welded structures based on the physical-data dual-drive strategy described in the present invention is:

[0069] Rautiainen et al. analyzed the fatigue behavior of the support column of the offshore cruise ship based on the finite element analysis method and measured the fatigue life of different positions. The fatigue load information of the support column structure of the offshore cruise ship provided by this study is: serving in a seawater corrosive environment, the ambient temperature is room temperature, the joint type is T-type / cross joint, the plate thickness is 5mm, 7mm, 15mm, the yield strength is 235MPa, the stress ratio is 0.1, the frequency is 2Hz, the weld state is full penetration, semi-penetration, and the load type is tension, compression + bending.

[0070] The average error of the fatigue life prediction results obtained based on the physical-data dual-driven method is 0.31, which is basically consistent with the accuracy of the fatigue life prediction of the DH36 steel welded hull structure in Specific Example 1, indicating that the predicted value obtained by the established architecture is consistent with the actual value, proving the good applicability of the proposed architecture in practical engineering applications such as offshore cruise ship support columns. Since the fatigue life prediction results obtained by actual measurement include some semi-penetration welds, and the fatigue life of semi-penetration welds is usually low. Therefore, the error fluctuation range of the prediction results obtained based on the established architecture is slightly increased (RMSE = 0.24), and the fatigue behavior of semi-penetration welds can be considered in the process of establishing the data set to further improve the accuracy of the method.

[0071] Example 3: Figure 4 As shown, the prediction result of the marine steel bridge welding structure obtained by the intelligent prediction method of fatigue life of engineering welded structure based on the physical-data dual-driven strategy of the present invention is:

[0072] Chen et al. investigated the fatigue performance of key fillet welds in marine steel bridge welded structures, and measured fatigue life under different weld toe morphologies and different stress ranges. The service information of the steel bridge welded structure in this study is: serving in seawater corrosion environment, joint type is T-type / cross joint, plate thickness is 12mm, 14mm, yield strength is 400MPa, stress ratio is 0.1, elastic modulus is 210GPa, Poisson's ratio is 0.33, the distance from the key position to the clamping position is 62-83mm, and the angle misalignment is 0-4.17°.

[0073] The average error of the fatigue life prediction results of the key corner weld positions in the marine steel bridge welded structure obtained based on the proposed prediction method is 0.34, which is basically consistent with the accuracy of the fatigue life prediction results of the marine engineering structure in specific examples 1 and 2, indicating that the proposed method has good prediction performance in the practical application of marine engineering structures. Different from the prediction results of the support column structure of the cruise ship at sea, since the marine steel bridge welded structure in this study does not contain key factors that cannot be considered in the data set, the fluctuation range of the prediction error is small (RMSE = 0.14), indicating that the proposed method can achieve reliable fatigue life prediction of marine engineering structures.

[0074] Example 4: Figure 5 As shown, the prediction result of the electric locomotive bogie structure obtained by the intelligent prediction method for fatigue life of engineering welded structures based on the physical-data dual-drive strategy of the present invention is:

[0075] He et al. studied the fatigue behavior of the electric locomotive bogie structure based on the structural stress method and predicted the fatigue life of the parts containing the bite edge defect. From this study, the fatigue load information of the electric locomotive bogie structure can be obtained as follows: the service environment is in the atmospheric environment, the ambient temperature is room temperature, the joint type is T-joint, the weld toe angle is 30-90°, the weld toe radius is 0.1-1mm, the stress concentration factor is 3.25-7.93, the yield strength is 345MPa, the elastic modulus is 206GPa, the plate thickness is 20mm, and the load type is tension and compression + bending.

[0076] In terms of fatigue life prediction of electric locomotive bogie structure, most of the prediction results obtained based on the physical-data dual-drive method are in good agreement with the structural stress method, with a 28% difference between the two prediction methods. The fatigue life prediction results obtained based on the proposed method are safer than those obtained by the structural stress method, thus providing stronger protection for the safe service of the bogie structure.

[0077] The intelligent prediction and analysis method of fatigue life of engineering welded structures based on the physical-data dual-drive strategy described in the present invention can capture and analyze important fatigue life influencing factors of engineering welded structures under different service conditions, and can comprehensively consider the impact of different influencing factors on fatigue life, thereby achieving fatigue life prediction with better adaptability. Based on the physical-data dual-drive strategy, the intelligent prediction method of fatigue life of engineering welded structures has good performance in direct prediction and indirect prediction of fatigue life values ​​and prediction of SN curves, and has good applicability in actual engineering welded structures. Thereby greatly reducing the large amount of manpower and material resources consumed by fatigue testing, providing new ideas and methods for fatigue performance analysis of major national engineering equipment and structures, and providing strong support for fatigue design of engineering welded structures.

[0078] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0079] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0080] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A fatigue life prediction method for welded structures based on a dual-drive strategy, characterized in that: The following steps are involved: Based on the deep convolutional neural network DCNN, the objective function model is generated by introducing physical constraints to realize the physical constraints of material strength, average stress, weld geometry, load conditions and size effect at the objective function level; According to the fatigue performance influencing factors of the engineering welded structure, a fatigue performance data set is obtained, model training is performed through the objective function model, and a fatigue life prediction model is constructed to predict the fatigue life of the engineering welded structure.

2. A method for predicting fatigue life of a welded structure based on a dual-drive strategy according to claim 1, characterized in that: In the process of constructing the target model, the target model is expressed as: In the formula, K t represents the stress concentration factor, E represents the elastic modulus, σ a represents the stress amplitude, Δσ represents the stress range, H c and h c They represent the cyclic hardening coefficient and cyclic hardening exponent respectively.

3. A method for predicting fatigue life of a welded structure based on a dual-drive strategy according to claim 2, characterized in that: In order to obtain the cyclic hardening index h c During the process, the cyclic hardening index h c It is expressed as: Where b and c represent the fatigue strength index and fatigue ductility index, respectively.

4. A method for predicting fatigue life of a welded structure based on a dual-drive strategy according to claim 3, characterized in that: In the acquisition process, it is expressed as: In the formula, σ′ f , ε′ f represent fatigue strength coefficient and fatigue ductility coefficient respectively.

5. A method for predicting fatigue life of a welded structure based on a dual-drive strategy according to claim 4, characterized in that: In the process of constructing the fatigue performance data set, an analysis is performed based on the service status of the engineering welded structure to obtain the factors affecting fatigue performance, construct the original fatigue performance data set, and generate the fatigue performance data set after data enhancement through the borderline synthetic minority class oversampling technology Borderline-SMOTE.

6. A method for predicting fatigue life of a welded structure based on a dual-drive strategy according to claim 5, characterized in that: In the process of constructing the original fatigue performance data set, based on the data set corresponding to the fatigue performance influencing factors, the original fatigue performance data set is generated after data processing through data cleaning, data preprocessing and data screening, wherein the data whose deviation between the fatigue life value and the value on the SN curve exceeds 5 times the average deviation of all data in the data set, and the data whose deviation between the fatigue life value and the life value on the SN curve exceeds 10 times are removed.

7. The method for predicting fatigue life of a welded structure based on a dual-drive strategy according to claim 1, characterized in that: In the process of model training, based on the fatigue performance data set, the importance of each physical and chemical influencing factor is analyzed through the extreme gradient boosting algorithm, sensitivity analysis method, Pearson correlation coefficient and maximum information coefficient, and the importance analysis result is used as the initial weight of the model for model training.

8. The method for predicting fatigue life of a welded structure based on a dual-drive strategy according to claim 1, characterized in that: In the process of constructing the fatigue performance data set, if the data volume is sufficient, the Borderline-SMOTE method is used to implement data enhancement; if the data volume is insufficient, the generative adversarial network or genetic algorithm method is used to increase the data volume.

9. The method for predicting fatigue life of a welded structure based on a dual-drive strategy according to claim 5, characterized in that: In the process of obtaining the factors affecting fatigue performance, based on different load forms, the load spectrum is converted into nominal stress to obtain the factors affecting fatigue performance.

10. A welding structure fatigue life prediction system based on a dual-drive strategy, characterized in that: The system is used to implement a welding structure fatigue life prediction method based on a dual-drive strategy as described in any one of claims 1 to 9, and the system includes: The model building module, based on the deep convolutional neural network DCNN, generates the objective function model by introducing physical constraints, which is used to realize the physical constraints of material strength, average stress, weld geometry, load conditions and size effect at the objective function level; A data acquisition module is used to obtain fatigue performance data sets based on the fatigue performance influencing factors of engineering welded structures; The prediction module is used to perform model training based on the objective function model and the fatigue performance data set to construct a fatigue life prediction model for predicting the fatigue life of the engineering welded structure.

Images