Method for predicting fatigue strength of metal sample based on internal defect characteristics

Abstract

The invention provides a method for predicting fatigue strength of a metal sample based on internal defect characteristics, which comprises the following steps of: establishing a scaling model containing random ellipsoid defects, and obtaining a stress concentration coefficient through finite element simulation; extracting defect geometric features to construct a data set, and training to R2 > = 0.85 by using a machine learning model; performing micro-CT scanning on an actual sample, fitting defect point cloud into an ellipsoid by adopting an MEE algorithm, and inputting the ellipsoid into a model to predict stress concentration; and combining a small amount of fatigue test correction material parameters, and integrating a defect-free S-N curve to construct a prediction model. According to the method, three-dimensional defect reconstruction, machine learning and parameter correction mechanisms are fused, and accurate fatigue strength prediction under different service life working conditions is realized with the minimum experimental quantity by quantifying relevance between defect characteristics and stress concentration. Compared with a traditional means, the method has the advantages of being low in cost, high in precision and high in generalization, and a reliable solution is provided for engineering component service life evaluation.

Description

Technical Field

[0001] The present invention belongs to the technical field of fatigue strength prediction of metal materials, and in particular relates to a method for predicting the fatigue strength of a metal sample based on internal defect characteristics. Background Art

[0002] Existing fatigue strength prediction models for metallic materials have achieved some success in most applications, but their accuracy remains limited in service environments dominated by complex loads or material defects. Some traditional models often employ simplified linear assumptions when addressing the relationship between surface roughness and fatigue limit, making it difficult to fully reflect the nonlinear effects of surface condition on fatigue performance. In reality, surface roughness is significantly affected by processes such as heat treatment and surface hardening, resulting in significant complexity and variability.

[0003] Furthermore, traditional models often assume that defects in materials are uniformly distributed. However, defects in real materials are often random, with significant variations in morphology, size, density, and spatial distribution. These defect characteristics have a significant impact on fatigue performance. For example, surface or near-surface defects are more likely to be the initiation source of fatigue cracks than deep-seated defects. The clustering of defects or their proximity to areas of stress concentration can also increase the risk of crack propagation, thereby reducing the fatigue life of the material.

[0004] With the deepening of our understanding of fatigue failure mechanisms, the three-dimensional morphology of defects and their spatial location within materials are recognized as key factors influencing fatigue performance. Different types of defects (such as holes, inclusions, and cracks) and their relative location within a component (such as proximity to the surface or areas of stress concentration) can significantly alter the local stress distribution, thereby affecting the initiation and propagation of fatigue cracks. Therefore, there is an urgent need to develop prediction methods that comprehensively consider the three-dimensional morphology and spatial distribution characteristics of defects and their relationship to load action to improve the accuracy and reliability of fatigue strength assessment.

[0005] In additive manufacturing, due to its layer-by-layer formation method, complex defects such as pores and lack of fusion are easily formed in the material. These defects may become the initiation sites of fatigue cracks during service, adversely affecting the fatigue life of the structure. Therefore, establishing a method that can accurately describe defect characteristics and achieve efficient fatigue strength prediction will not only facilitate additive manufacturing quality control and process optimization, but also improve material service safety, reduce costs, and promote the engineering and sustainable development of this technology. Summary of the Invention

[0006] To address the above problems, the present invention discloses a method for predicting the fatigue strength of metal specimens based on internal defect characteristics. By combining finite element simulation with defect feature quantification, machine learning is used to establish a defect-stress response relationship, and combined with micro-CT three-dimensional defect reconstruction technology, a highly robust defect-fatigue strength mapping model is constructed by optimizing defect feature extraction and model generalization capabilities.

[0007] To achieve the above object, the technical solution of the present invention is as follows:

[0008] A method for predicting fatigue strength of a metal specimen based on internal defect characteristics comprises the following steps:

[0009] S1. According to the sample morphology and structural information, a proportional scaling model is constructed in the finite element analysis software, and randomly distributed unequal ellipsoid defects are preset inside. Combined with the actual stress conditions, simulation analysis is performed and the stress concentration factor (K t ); extract key features of each anisometric ellipsoid: construct a data set;

[0010] S2. Based on the data set, the model is trained to establish the defect-stress concentration factor (K t ) mapping relationship, when the determination coefficient R 2 The model converges when ≥0.85;

[0011] S3. Perform micro-CT scanning on the actual sample, extract the internal defect point cloud information, and use the MEE algorithm to fit it into an isometric ellipsoid and extract the feature value;

[0012] S4. Input the trained model in S2 to obtain the predicted stress concentration factor; select any fatigue test results to correct the material parameters, integrate the defect-free SN curve under the stress condition and the corrected parameters to reconstruct the expression, and finally establish a high-precision fatigue strength prediction expression.

[0013] Furthermore, the scaling model is constructed in the finite element analysis software as described in step S1, specifically as follows:

[0014] A three-dimensional model of a dog-bone fatigue specimen was established in the Abaqus finite element software, and 10 to 15 isometric ellipsoidal defects were randomly preset inside the model. The lengths of the three semi-axes of each ellipsoidal defect were randomly generated in the range of 20 μm to 500 μm, and the spatial orientations of the three semi-axes were randomly distributed to accurately simulate the defect characteristics in actual metal materials.

[0015] Furthermore, the finite element analysis software described in step S1 is Abaqus, and the corresponding load conditions and material parameters are set in the analysis software: the Young's modulus of the material is set to 110GPa, the Poisson's ratio is set to 0.33, and the loading mode is tension-tension cyclic stress; finite element simulation calculations are performed according to the set conditions to obtain the stress distribution of the defect area in the model. The maximum von Mises stress value near the defect and the nominal stress value applied by the model are extracted to calculate the stress concentration factor (K t ), which is defined as the ratio of the maximum von Mises stress to the nominal stress.

[0016] Furthermore, there are four key features in step S1, namely:

[0017] (1) The maximum aspect ratio of the projected ellipse of the plane family in the force direction, that is, This characteristic quantity reflects the projection change of the defect in different stress directions and can capture the geometric response characteristics of the defect producing stress concentration tendency along the principal stress direction.

[0018] (2) The angle between the major axis of the projected ellipse with the maximum aspect ratio and the vertical plane of force, i.e., θ proj,max This angle feature simultaneously regulates the three core mechanisms of projected area, stress concentration and anisotropy, and can quantify the spatial relationship between the defect principal axis direction and the load direction, which helps to identify potential crack initiation paths.

[0019] (3) The shortest distance from the ellipsoid to the surface. This distance not only affects the stress distribution in the defect area, but also determines the effect of the defect on the overall mechanical properties of the material to a certain extent. In particular, when the defect distance to the surface is less than 0.5, the stress concentration factor increases sharply.

[0020] (4) The square root of the projected area in the direction of force. This characteristic quantity effectively considers the size of the defect and the influence of the force direction.

[0021] Furthermore, step S3 is specifically as follows:

[0022] Micro-CT scanning was performed on the actual metal sample to obtain the 3D point cloud data of its internal defects. Gaussian filtering was used for noise reduction, and the voxel size was set to 5.6×5.6×5.6μm. 3 ; Use image reconstruction and defect segmentation algorithms to extract point cloud information of independent defect bodies, and use the minimum ellipsoid envelope (MEE) algorithm to fit each defect to obtain the geometric model of the isometric ellipsoid; after fitting, extract the ellipsoid feature values according to the aforementioned method, and import them as input data into the trained machine learning model to predict the stress concentration coefficient corresponding to each defect, thereby realizing a rapid evaluation of the actual defect stress response.

[0023] Furthermore, step S4 is specifically as follows:

[0024] (S41) The fatigue life data of the material in the presence of defects is obtained through fatigue testing, and the Basquin formula is used as the basic expression, and the stress concentration factor K is introduced. t Make corrections, and the correction expression is as follows

[0025] σ max =3684.74·(K t ) -0.0938 ·(N f ) -0.1251

[0026] where σ max To predict fatigue strength, K t is the stress concentration factor, N f is the expected fatigue life result.

[0027] (S42) Integrate the defect-free SN curve under this stress condition and the modified parameter reconstruction expression to finally establish a high-precision fatigue strength prediction expression

[0028]

[0029] According to the Murakami formula, the correction factor C is determined by the defect location: when the defect is located on the contact surface, C = 1.41; when the defect is close to the surface, C = 1.43; and when the defect is located inside the material, C = 1.56. Where N is the expected fatigue life result, N lim is the fatigue limit life of the material, b is the coefficient of the Basquin equation, is the equivalent length of the internal defect projection area, and HV is the Vickers hardness.

[0030] The beneficial effects of the present invention are:

[0031] (1) A new defect equivalent modeling method is proposed: This invention uses micro-CT scanning combined with the minimum ellipsoid envelope (MEE) algorithm to fit irregular defects into an isometric ellipsoid, significantly improving the characterization capability of real complex defects and making defect morphology modeling more engineering operable and versatile. Different from the traditional method of using the defect equivalent diameter as a single descriptive parameter, this invention establishes four key defect characteristic dimension, which more comprehensively characterizes the relationship between the defect shape, position and loading direction, and enhances the model's ability to express defect heterogeneity.

[0032] (2) The maximum aspect ratio of the projected ellipse of the plane family in the stress direction is proposed as one of the core characteristic quantities. This characteristic quantity reflects the projection change of the defect in different stress directions and can capture the geometric response characteristics of the stress concentration trend generated by the defect along the principal stress direction, significantly improving the sensitivity and accuracy of the stress concentration factor prediction.

[0033] (3) The angle between the major axis of the maximum aspect ratio projected ellipse and the vertical plane of the load is introduced as a characterization parameter. This angle feature can quantify the spatial relationship between the main axis direction of the defect and the load direction, which helps to identify potential crack initiation paths and plays an important role in establishing the association between defects and fatigue response.

[0034] (4) A defect-stress response prediction model based on machine learning was constructed. By inputting the above-mentioned key feature quantities into the training model, a high-precision mapping between defect parameters and stress concentration factors was achieved, which significantly improved the generalization ability of fatigue strength prediction.

[0035] (5) An efficient modeling framework suitable for small sample prediction. The method of the present invention can realize fatigue performance prediction with only a small amount of measured data without relying on a large number of fatigue tests, which effectively reduces the cost of engineering testing and improves the industrial application feasibility of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic flow chart of the present invention;

[0037] Figure 2 1 is a schematic diagram of fatigue specimen and defect equivalence in an embodiment of the present invention;

[0038] Figure 3 Schematic diagram of defect characteristic quantities in an embodiment of the present invention;

[0039] Figure 4 1 is a micro-CT scan result of a defect and a defect fitting schematic diagram in an embodiment of the present invention;

[0040] Figure 5 is a comparison chart of the prediction results of the model proposed in the embodiment of the present invention and the traditional model on various test samples;

[0041] Figure 6 3 is a comparison chart of the mean prediction error (MAPE) between the model proposed in the embodiment of the present invention and the traditional model. DETAILED DESCRIPTION

[0042] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0043] The present invention provides a method for predicting the fatigue strength of a metal specimen based on internal defect characteristics. The specific steps and principles are further explained below with reference to an embodiment:

[0044] like Figure 2 The fatigue specimen shown is a reference. This specimen has a dumbbell-shaped structure, consisting of cylindrical segments on both ends connected by a circular arc transition section with a radius of R20 mm. The left and right sides are straight cylindrical segments with a diameter of Φ12 mm, and the transition section is designed as a circular arc surface with a radius of 20 mm. A dogbone fatigue specimen was constructed in SolidWorks at a moderate scale. Multiple anisometric ellipsoids were programmatically constructed within the model and then merged with the fatigue specimen using a Boolean algorithm. The anisometric ellipsoids were randomly distributed in position, size, and spatial orientation (rotation angle).

[0045] Import the constructed equivalent digital model of the fatigue specimen with defects into the finite element analysis software (Abaqus is used in this example), and set the corresponding load conditions and material parameters in the Abaqus finite element analysis software. In this embodiment, the Young's modulus of the material is set to 110GPa, the Poisson's ratio is 0.33, and the loading mode is tension-tension cyclic tensile stress. Perform finite element simulation calculations according to the set conditions to obtain the stress distribution of the defect area in the model. Extract the maximum von Mises stress value near the defect and the nominal stress value applied by the model, and calculate the stress concentration factor (K t ), which is defined as the ratio of the maximum von Mises stress to the nominal stress.

[0046] like Figure 3 As shown in the figure, the four key characteristic quantities of each defect are extracted, namely: (1) the maximum aspect ratio of the projected ellipse of the plane family in the force direction, that is, (2) The angle between the major axis of the projected ellipse with the maximum aspect ratio and the vertical plane of force, i.e., θ proj,max , (3) the shortest distance from the ellipsoid to the surface, and (4) the square root of the projected area in the direction of force. A training data set is constructed based on the above feature quantities, and the number of training samples is no less than 500 groups.

[0047] The extracted key defect characteristics and the corresponding stress concentration factor are used as input data and imported into the machine learning algorithm for model training. During the training process, the fitting effect of the model is continuously evaluated. When the coefficient of determination (R 2 ) reaches no less than 0.85, the model training is determined to be complete and has convergence and prediction capabilities. In this embodiment, the XGBoost regression algorithm is used for modeling training. The determination coefficient of the final model is 0.88, which meets the accuracy requirements and can be used for subsequent defect stress response prediction.

[0048] like Figure 4As shown in the figure, a micro-CT scan is performed on an actual metal sample to obtain 3D point cloud data of its internal defects (Avizo software is used in this example). In this embodiment, the micro-CT scan resolution is 5.6 μm (the specific resolution depends on the material), and Gaussian filtering is used for noise reduction during the reconstruction process. The voxel size is set to 5.6×5.6×5.6 μm. 3 Image reconstruction and defect segmentation algorithms are used to extract point cloud information of independent defects, and the minimum ellipsoid envelope (MEE) algorithm is used to fit each defect to obtain a geometric model of an isometric ellipsoid. After fitting, the ellipsoid features are extracted according to the aforementioned method and imported as input data into the trained machine learning model to predict the stress concentration factor corresponding to each defect, thereby achieving a rapid assessment of the actual defect stress response.

[0049] The fatigue life data of the material under the condition of defects is obtained through fatigue testing, and combined with the defect-free SN curve of the material under the same working conditions, a fatigue strength prediction model based on physical constraints is constructed. In this embodiment, the Basquin formula is used as the basic expression, and the stress concentration factor K is introduced. t Correction is made to establish a fatigue strength prediction formula that includes the effect of defects. The final corrected expression is as follows: max =3684.74·(K t ) -0.0938 ·(N f ) -0.1251 . Where σ max To predict fatigue strength, K t is the stress concentration factor, N f The fatigue life results are as follows: Figure 5 and Figure 6 As shown, ML is the model proposed in the present invention, and Murakami is the traditional fatigue strength prediction model selected for comparison in this embodiment. According to the Murakami formula, the value of the correction factor C is determined by the defect location: when the defect is located on the contact surface, C = 1.41; when the defect is close to the surface, C = 1.43; and when the defect is located inside the material, C = 1.56. Among them, the fatigue limit life of the material is recorded as N lim In this study, the fatigue limit life of TC11 titanium alloy is defined as 10 7 cycles; the Basquin equation coefficient b is obtained by fitting the material parameters of the SN curve of the defect-free sample. The value determined in this embodiment is -0.125, and the HV is 390. Figure 6 The vertical axis is the mean absolute percentage error (MAPE), which is used to quantitatively evaluate the accuracy of each model in the fatigue strength prediction process. Figure 6The proposed ML model demonstrates superior prediction accuracy in test samples containing actual defects, with an average error of only 4.06%, far exceeding the 39.32% of the Murakami model. These results demonstrate that the proposed model significantly improves the accuracy and stability of fatigue strength prediction by fully considering the three-dimensional morphology and spatial characteristics of defects, validating the effectiveness and engineering applicability of the proposed method in defect-dominated fatigue scenarios.

[0050] It should be noted that the above content merely illustrates the technical idea of the present invention and cannot be used to limit the scope of protection of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications all fall within the scope of protection of the claims of the present invention.

Claims

1. A method for predicting fatigue strength of metal specimens based on internal defect characteristics, characterized by: The following steps are involved: S1. According to the sample morphology and structural information, a proportional scaling model is constructed in the finite element analysis software, and randomly distributed unequal ellipsoid defects are preset inside. Combined with the actual stress conditions, simulation analysis is performed and the stress concentration factor (K t ); extract key features of each anisometric ellipsoid: construct a data set; S2. Based on the data set, the model is trained to establish the defect-stress concentration factor (K t ) mapping relationship, when the determination coefficient R 2 The model converges when ≥0.85; S3. Perform micro-CT scanning on the actual sample, extract the internal defect point cloud information, and use the MEE algorithm to fit it into an isometric ellipsoid and extract the feature value; S4. Input the trained model in S2 to obtain the predicted stress concentration factor; Any fatigue test results are selected to correct the material parameters, and the defect-free SN curve under the stress condition is integrated with the corrected parameters to reconstruct the expression, and finally a high-precision fatigue strength prediction expression is established.

2. The method for predicting fatigue strength of a metal specimen based on internal defect characteristics according to claim 1, characterized in that: The construction of the proportional scaling model in the finite element analysis software described in step S1 is as follows: A three-dimensional model of a dog-bone fatigue specimen was established in the Abaqus finite element software, and 10 to 15 isometric ellipsoidal defects were randomly preset inside the model. The lengths of the three semi-axes of each ellipsoidal defect were randomly generated in the range of 20 μm to 500 μm, and the spatial orientations of the three semi-axes were randomly distributed to accurately simulate the defect characteristics in actual metal materials.

3. The method for predicting fatigue strength of a metal specimen based on internal defect characteristics according to claim 1, characterized in that: The finite element analysis software described in step S1 is Abaqus. The corresponding load conditions and material parameters are set in the analysis software: the Young's modulus of the material is set to 110 GPa, the Poisson's ratio is set to 0.33, and the loading mode is tension-tension cyclic stress. Finite element simulation calculations are performed according to the set conditions to obtain the stress distribution of the defect area in the model; Extract the maximum von Mises stress value near the defect and the nominal stress value applied by the model, and calculate the stress concentration factor (K t ), which is defined as the ratio of the maximum vonMises stress to the nominal stress.

4. The method for predicting fatigue strength of a metal specimen based on internal defect characteristics according to claim 1, characterized in that: There are four key features in step S1, namely: (1) The maximum aspect ratio of the projected ellipse of the plane family in the force direction, that is, This characteristic reflects the projection change of the defect in different stress directions and can capture the geometric response characteristics of the defect producing stress concentration tendency along the principal stress direction. (2) The angle between the major axis of the projected ellipse with the maximum aspect ratio and the vertical plane of force, i.e., θ proj,max This angle feature simultaneously regulates the three core mechanisms of projected area, stress concentration, and anisotropy, and can quantify the spatial relationship between the defect principal axis direction and the load direction, which helps to identify potential crack initiation paths; (3) The shortest distance from the ellipsoid to the surface. This distance not only affects the stress distribution in the defect area, but also determines the effect of the defect on the overall mechanical properties of the material to a certain extent. In particular, when the defect distance to the surface is less than 0.5, the stress concentration factor increases sharply. (4) The square root of the projected area in the direction of force. This characteristic quantity effectively considers the size of the defect and the influence of the force direction.

5. The method for predicting fatigue strength of a metal specimen based on internal defect characteristics according to claim 1, characterized in that: Step S3 is as follows: Micro-CT scanning was performed on the actual metal sample to obtain the 3D point cloud data of its internal defects. Gaussian filtering was used for noise reduction, and the voxel size was set to 5.6×5.6×5.6μm. 3 ; Use image reconstruction and defect segmentation algorithms to extract point cloud information of independent defect bodies, and use the minimum ellipsoid envelope algorithm to fit each defect to obtain the geometric model of the isometric ellipsoid; after fitting, extract the ellipsoid feature values according to the aforementioned method, and import them as input data into the trained machine learning model to predict the stress concentration coefficient corresponding to each defect, thereby realizing a rapid evaluation of the actual defect stress response.

6. The method for predicting fatigue strength of a metal specimen based on internal defect characteristics according to claim 1, characterized in that: Step S4 is specifically as follows: (S41) The fatigue life data of the material in the presence of defects is obtained through fatigue testing, and the Bakins formula is used as the basic expression, and the stress concentration factor (K t ) is corrected, and the corrected expression is as follows s max =3684.74·(K t ) -0.0938 ·(N f ) -0.1251 where σ max To predict fatigue strength, K t is the stress concentration factor, N f is the expected fatigue life result; (S42) Integrate the defect-free SN curve under this stress condition and the modified parameter reconstruction expression to finally establish a high-precision fatigue strength prediction expression According to the Murakami formula, the correction factor C is determined by the defect location: when the defect is located on the contact surface, C = 1.41; when the defect is close to the surface, C = 1.43; and when the defect is located inside the material, C = 1.

56. Where N is the expected fatigue life result, N lim is the fatigue limit life of the material, b is the coefficient of the Basquin equation, is the equivalent length of the internal defect projection area, and HV is the Vickers hardness.

Images