A machine learning-based passenger car frame weld fatigue simulation calculation method

Abstract

The present application belongs to the technical field of fatigue life simulation, and provides a bus skeleton weld fatigue simulation calculation method based on machine learning, comprising: using a fatigue testing machine to perform fatigue test on an automobile thin-walled part, obtaining fatigue life data under different stress intensities according to weld types, and dividing a training set and a test set; adopting a support vector machine algorithm to establish a weld fatigue life prediction model with the training set; within the stress range of the workpiece, obtaining fatigue life prediction values corresponding to each refined stress level point based on the weld fatigue life prediction model to form a high-density prediction data set; performing linear regression fitting on the high-density prediction data set in a logarithmic coordinate system to construct a stress-life curve. The present application expands the original sparse test data points into a high-density prediction data set through the SVR algorithm, and then performs linear regression fitting on the S-N curve obtained, which can improve the reliability of the simulation calculation result.

Description

Technical Field

[0001] This invention relates to the field of fatigue life simulation technology, specifically to a fatigue simulation calculation method for bus frame welds based on machine learning. Background Technology

[0002] The bus frame is a load-bearing structure made of a large number of thin-walled steel sections welded together. Due to stress concentration, residual welding stress, and changes in material microstructure, the weld area is a high-risk location for fatigue failure. Accurately assessing the fatigue life of the welds is of significant engineering importance for ensuring bus safety and reducing maintenance costs. Currently, the fatigue life assessment of welded structures is usually done by conducting fatigue life tests on specimens to obtain the SN curve. This involves measuring the stress amplitude experienced by the test specimen. As the ordinate of the curve, or the maximum stress Using fatigue life, i.e., the number of stress cycles, as the vertical axis, Use the x-axis to plot the standard fatigue life curve.

[0003] Considering the limited experimental costs, small sample size, and significant impact of experimental errors, as well as the production inconsistencies in the welding process of thin-walled automotive parts that affect the accuracy of the plotted SN curves, the material properties in the fatigue simulation software's material library cannot accurately reflect the material properties during production and processing, thus increasing the risk to design reliability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a machine learning-based method for simulating fatigue in bus frame welds, thereby solving the problem of poor accuracy of the SN curves generated by the small sample size in current weld fatigue tests.

[0005] This invention provides a machine learning-based method for simulating fatigue in bus frame welds, comprising: S1. Use a fatigue testing machine to conduct fatigue tests on thin-walled automotive parts, obtain fatigue life data under different stress intensities according to weld type, and divide the data into training set and test set; weld types include 2D co-node welds, software-defined welds, and 3D solid welds; S2. A support vector machine algorithm is used to establish a weld fatigue life prediction model using the training set; S3. Within the stress range of the workpiece, refine the stress level points with a density of not less than ten to twenty times the original test stress interval, and obtain the fatigue life prediction value corresponding to each refined stress level point based on the weld fatigue life prediction model to form a high-density prediction dataset. S4. Perform linear regression fitting on the high-density prediction dataset in a logarithmic coordinate system to construct a stress-life curve; S5. Import the stress-life curve into the material property library of the fatigue simulation software, and perform fatigue simulation calculations on the bus frame welds based on the imported material properties.

[0006] Optionally, fatigue life data includes a sample pair ,in For the first The number of cycles experienced by each sample For the first The stress intensity experienced by each sample; After that and All sets were normalized and divided into training and test sets according to a preset ratio; the normalization formula is as follows: , ∈[-1,1] represents the normalized data.

[0007] Optionally, S2 includes: Using Gaussian kernel function and cross-validation, we can find the penalty factor C and the variance of Gaussian kernel function. And, in conjunction with the training set, establish a weld fatigue life prediction model; The trained weld fatigue life prediction model is evaluated using mean square error and coefficient of determination. If both mean square error and coefficient of determination meet the corresponding conditions, the weld fatigue life prediction model is output; otherwise, the model parameters are modified.

[0008] Optionally, the Gaussian kernel function is: ,parameter Let be the variance of the Gaussian kernel function; The weld fatigue life prediction model is as follows: , in, for: , The insensitivity coefficient, , This is the optimal solution to the dual form problem of the Lagrange function; The number of support vectors, i.e. Correspondence The number of samples is the number of sample pairs that fall on and outside the two boundaries of the SVR model.

[0009] Optionally, the high-density prediction dataset is obtained by taking the logarithm of the fatigue life prediction values ​​at different refined stress levels, specifically: Substituting the refined stress level points into the weld fatigue life prediction model yields n model prediction points. ; After inversely normalizing the model's predicted points and taking the logarithm, n fitted points are obtained. ; Anomaly jump detection is performed on the fitted points, and fitted points whose rate of change with adjacent points is greater than a threshold are removed.

[0010] Optionally, S4 includes: From n fitting points The stress-life curve equation obtained from linear regression: ; in, , , , For n fitting points The mean of the x and y coordinates.

[0011] Optionally, S5 includes: Establish a finite element model of the bus frame; By associating weld types with the finite element model of the bus frame, a weld fatigue simulation model is established. Based on the type of load spectrum and the constraints, strength software was used to calculate the strength of the bus frame weld under unit excitation, and the strength result file was output. Import the stress-life curves and strength result files into the fatigue simulation software to perform fatigue simulation calculations on the weld seams of the bus frame.

[0012] By adopting the above technical solution, this application has the following beneficial effects: This invention generates high-density prediction data points by modeling at refined stress levels, expands the original sparse experimental data points into a high-density prediction dataset using the SVR algorithm, and fits the obtained SN curve by linear regression in a logarithmic coordinate system. Since fatigue experiments on metallic materials are usually time-consuming and therefore have a limited number of samples, this invention improves the reliability of fatigue simulation calculations with small samples compared to directly fitting curves from the original sparse data points, reduces reliance on a large number of fatigue experiments, and lowers experimental costs and time. Attached Figure Description

[0013] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0014] Figure 1 The flowchart illustrates a machine learning-based fatigue simulation calculation method for bus frame welds provided in an embodiment of the present invention. Figure 2A schematic diagram of the finite element model of the bus frame provided in an embodiment of the present invention is shown; Figure 3 A schematic diagram of establishing a fatigue simulation weld provided by an embodiment of the present invention is shown; Figure 4 A schematic diagram of importing a finite element model for fatigue analysis provided in an embodiment of the present invention is shown; Figure 5 The diagram shows the fatigue life cloud map of the bus frame weld provided in an embodiment of the present invention; Figure 6 The figure shown is a fitting simulation diagram provided by an embodiment of the present invention. Detailed Implementation

[0015] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of the present invention and are therefore merely examples, and should not be construed as limiting the scope of protection of the present invention.

[0016] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0017] In one embodiment, such as Figure 1 As shown, a machine learning-based method for simulating fatigue in bus frame welds is provided, including: S1. Use a fatigue testing machine to conduct fatigue tests on thin-walled automotive parts, obtain fatigue life data under different stress intensities according to weld type, and divide the data into training set and test set; weld types include 2D common node welds, software-defined welds, and 3D solid welds.

[0018] Fatigue life data includes a sample pair ,in For the first The number of cycles experienced by each sample For the first The stress intensity experienced by each sample; After that and All sets were normalized and divided into training and test sets according to a preset ratio; the normalization formula is as follows: , ∈[-1,1] represents the normalized data.

[0019] S2. A support vector machine algorithm is used to establish a weld fatigue life prediction model using the training set.

[0020] Optionally, S2 includes: S21. Use the Gaussian kernel function and cross-validation to find the penalty factor C and the variance of the Gaussian kernel function. A weld fatigue life prediction model was established by combining the training set.

[0021] The interpolated portion of the high-density prediction dataset output by the weld prediction model can only be used as dummy samples. These dummy samples need to be evaluated for usability before being used for linear fitting to obtain the SN curve. Hold-out validation or cross-validation can be used to verify the prediction bias in the dummy sample region using partial real experimental data. If the mean squared error and coefficient of determination still meet the requirements in the densely populated dummy sample region, the dummy samples are usable; otherwise, samples in that region should be removed or corrected.

[0022] S22. Evaluate the trained weld fatigue life prediction model using the mean square error and the coefficient of determination. If both the mean square error and the coefficient of determination meet the corresponding conditions, output the weld fatigue life prediction model; otherwise, modify the model parameters.

[0023] If the mean squared error and coefficient of determination cannot meet the corresponding conditions, the model parameters can be modified by using grid search, random search, or Bayesian optimization methods to re-search for C and D within the preset parameter space. The optimal combination is then determined, and cross-validation is performed again. If the conditions are still not met, consider adjusting the kernel function type or increasing the amount of data. The risk of secondary training is that it may overfit the validation set; therefore, a separate test set should be used for final validation.

[0024] The Gaussian kernel function is: ,parameter Let be the variance of the Gaussian kernel function; The fatigue life prediction model for welds is as follows: , in, for: , The insensitivity coefficient, , This is the optimal solution to the dual form problem of the Lagrange function; The number of support vectors, i.e. Correspondence The number of samples is the number of sample pairs that fall on and outside the two boundaries of the SVR model.

[0025] S3. Within the stress range of the workpiece, refine the stress level points with a density of not less than ten to twenty times the original test stress interval. Based on the weld fatigue life prediction model, obtain the fatigue life prediction value corresponding to each refined stress level point to form a high-density prediction dataset.

[0026] The high-density prediction dataset is obtained by taking the logarithm of the fatigue life prediction values ​​at different refined stress levels, specifically: T1. Substitute the refined stress level points into the weld fatigue life prediction model to obtain n model prediction points. .

[0027] T2. After denormalizing the model's predicted points, take the logarithm to obtain n fitted points. .

[0028] T3. Perform abnormal jump detection on the fitted points and filter out fitted points whose rate of change with neighboring points exceeds a threshold. Since the interpolation part in the high-density prediction dataset is essentially the model's prediction output at the refined stress points, it belongs to the model's inference result rather than the original sample; step T3 performs quality pre-detection on the interpolation points output by the model, and ensures data quality by filtering out fitted points with abnormal jumps.

[0029] S4. Perform linear regression fitting on the high-density prediction dataset in a logarithmic coordinate system to construct the stress-life curve.

[0030] From n fitting points The stress-life curve equation obtained from linear regression: ; in, , , , For n fitting points The mean of the x and y coordinates.

[0031] S5. Import the stress-life curve into the material property library of the fatigue simulation software, and perform fatigue simulation calculations on the bus frame welds based on the imported material properties.

[0032] In one embodiment, S5 includes: S51. Establish a finite element model of the bus frame.

[0033] S52. Associate the weld type on the finite element model of the bus frame and establish a weld fatigue simulation model.

[0034] S53. Based on the type of load spectrum and the constraint conditions, use strength software to calculate the strength of the bus frame weld under unit excitation and output the strength result file.

[0035] S54. Import the stress-life curve and strength result file into the fatigue simulation software to perform fatigue simulation calculations on the bus frame welds.

[0036] The following is a specific scenario to illustrate the operation process of step S5. The specific steps are as follows: 1) Establish a finite element model of the bus frame. Fatigue analysis of the bus welded structure requires the establishment of a corresponding finite element model. It is recommended that the element size for sheet metal parts be between 5mm and 15mm, and the element size for solid parts be 5mm. For smaller or more complex components, the dimensions can be appropriately reduced. After establishing the finite element model of the bus welded structure, perform modal analysis to check the correctness of the model assembly. If any abnormalities are found in the modal modes, the model connections need to be modified until the model assembly is confirmed to be correct. Figure 2 As shown.

[0037] 2) Establish a weld fatigue simulation model on the finite element model of the bus frame. Weld types include 2D common-node welds, software-defined welds, and 3D solid welds. Establish appropriate weld types as needed, such as... Figure 3 As shown.

[0038] 3) Based on the type of load spectrum and the constraint conditions, strength software is used to calculate the strength of the bus frame weld under unit excitation, and the strength calculation result file *op2 is output.

[0039] 4) Import the fatigue analysis finite element model into the software, such as... Figure 4 As shown.

[0040] In the software, the imported model is divided into different groups based on different materials. Materials are created and called, with corresponding parameters entered in the material properties, and SN curves imported. Load spectra and constraints are defined, and the strength result file *.op2 under unit excitation is imported. Welding quality and fatigue simulation influence parameters are set, as are analysis parameters and output requirements. Fatigue simulation calculations of the bus frame welds are performed, and the fatigue life cloud map of the bus frame welds is output, such as... Figure 5 As shown.

[0041] like Figure 6 As shown, the experimental data exhibits some dispersion, but the overall experimental data conforms to statistical laws and can be used to provide data reference for the fatigue strength of welded joints.

[0042] Specific analysis revealed that the S-curve fatigue SN curves of the stainless steel structure showed high fitting accuracy, with an R-squared value of 0.99 for the S-curve, a linear R-squared value of 0.977 for the log-log scale, and highly significant F-tests for both equations. The new sample from Australian steel also exhibited high fitting accuracy for its structural fatigue SN curves, with an R-squared value of 0.929 for the S-curve, a linear R-squared value of 0.991 for the log-log scale, and highly significant F-tests for both equations. The obtained SN curve equations can be imported into the material library of fatigue simulation software.

[0043] The above embodiments are only used to provide a detailed description of the technical solutions of this application. However, the descriptions of the above embodiments are only for the purpose of helping to understand the methods of the embodiments of the present invention and should not be construed as limiting the embodiments of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art should be covered within the protection scope of the embodiments of the present invention.

Claims

1. A method for fatigue simulation calculation of bus frame welds based on machine learning, characterized in that, include: S1. Use a fatigue testing machine to conduct fatigue tests on thin-walled automotive parts, obtain fatigue life data under different stress intensities according to weld type, and divide the data into training set and test set; weld types include 2D co-node welds, software-defined welds, and 3D solid welds; S2. A support vector machine algorithm is used to establish a weld fatigue life prediction model using the training set; S3. Within the stress range of the workpiece, refine the stress level points with a density of not less than ten to twenty times the original test stress interval, and obtain the fatigue life prediction value corresponding to each refined stress level point based on the weld fatigue life prediction model to form a high-density prediction dataset. S4. Perform linear regression fitting on the high-density prediction dataset in a logarithmic coordinate system to construct a stress-life curve; S5. Import the stress-life curve into the material property library of the fatigue simulation software, and perform fatigue simulation calculations on the bus frame welds based on the imported material properties.

2. The method according to claim 1, characterized in that, Fatigue life data includes a sample pair ,in For the first The number of cycles experienced by each sample For the first The stress intensity experienced by each sample; After that and All data were normalized and divided into training and testing sets according to a preset ratio. The normalization formula is: , ∈[-1,1] represents the normalized data.

3. The method according to claim 2, characterized in that, S2 include: Using Gaussian kernel function and cross-validation, we can find the penalty factor C and the variance of Gaussian kernel function. And, in conjunction with the training set, establish a weld fatigue life prediction model; The trained weld fatigue life prediction model is evaluated using the mean square error and the coefficient of determination. If both the mean square error and the coefficient of determination meet the corresponding conditions, the weld fatigue life prediction model is output; otherwise, the model parameters are modified.

4. The method according to claim 3, characterized in that, The Gaussian kernel function is: ,parameter Let be the variance of the Gaussian kernel function; The weld fatigue life prediction model is as follows: ， in, for: ， The insensitivity coefficient, , This is the optimal solution to the dual form problem of the Lagrange function; The number of support vectors, i.e. Correspondence The number of samples is the number of sample pairs that fall on and outside the two boundaries of the SVR model.

5. The method according to claim 4, characterized in that, The high-density prediction dataset is obtained by taking the logarithm of the fatigue life prediction values ​​at different refined stress levels, specifically: Substituting the refined stress level points into the weld fatigue life prediction model yields n model prediction points. ; After denormalizing the model's predicted points and taking the logarithm, n fitted points are obtained. ; Anomaly jump detection is performed on the fitted points, and fitted points whose rate of change with adjacent points is greater than a threshold are removed.

6. The method according to claim 5, characterized in that, S4 include: From n fitting points The stress-life curve equation obtained from linear regression: ; in, , , , For n fitting points The mean of the x and y coordinates.

7. The method according to claim 1, characterized in that, S5 include: Establish a finite element model of the bus frame; By associating weld types with the finite element model of the bus frame, a weld fatigue simulation model is established. Based on the type of load spectrum and the constraints, strength software was used to calculate the strength of the bus frame weld under unit excitation, and the strength result file was output. Import the stress-life curves and strength result files into the fatigue simulation software to perform fatigue simulation calculations on the weld seams of the bus frame.

Images