Method for predicting high-cycle fatigue life of titanium alloy after multi-step thermal deformation based on machine learning

Abstract

The invention discloses a method for predicting the high-cycle fatigue life of a titanium alloy after multi-step thermal deformation based on machine learning, and belongs to the technical field of titanium alloy fatigue life prediction.The method comprises the steps that 1, a plurality of samples are cut from a Ti-6Al-4V titanium alloy after diffusion bonding to be subjected to a multi-step thermal deformation test, and fatigue samples are prepared; 2, performing high-cycle fatigue experiments under different loading conditions to obtain high-cycle fatigue data of the material; 3, performing normalization processing on the data, dividing a data set into a training set and a test set, and improving the total data amount by adopting data enhancement; 4, obtaining a machine learning model based on a training set learning result; and 5, evaluating the model by adopting a determination coefficient R2, an average absolute relative error AARE, a root mean square error RMSE and an average absolute error MAE to finish model establishment, and predicting the high-cycle fatigue life of the titanium alloy after multi-step thermal deformation by utilizing the model. According to the method, the high-cycle fatigue life of the Ti-6Al-4V titanium alloy after multi-step thermal deformation is accurately predicted through a machine learning model.

Description

Technical Field

[0001] This invention relates to the field of fatigue life prediction technology for titanium alloys, and in particular to a machine learning-based method for predicting high-cycle fatigue life of titanium alloys after multi-step hot deformation. Background Technology

[0002] Titanium alloys are widely used in the aerospace field due to their high strength, high temperature resistance, and strong resistance to foreign object damage. Many parts requiring lightweight, high strength, and high-temperature operation are now manufactured using titanium alloys.

[0003] Titanium alloys can be classified into three types based on their phase composition: α-type, β-type, and α+β-type. Compared to α-type and β-type alloys, components made from α+β-type titanium alloys can reduce weight by approximately 9%. Therefore, Ti-6Al-4V titanium alloy is currently the most widely used α+β-type titanium alloy. The microstructure of Ti-6Al-4V titanium alloy changes before and after hot forming, resulting in a significant difference in high-cycle fatigue life compared to the original material. Therefore, accurate prediction of the high-cycle fatigue life of titanium alloys after forming is necessary to improve component reliability.

[0004] The forming process of Ti-6Al-4V titanium alloy involves complex thermo-mechanical coupling, making it difficult to obtain the desired part through single-step forming. Furthermore, the microstructure undergoes significant changes during multi-step forming, thus affecting the material's high-cycle fatigue life. Therefore, in actual production, studying the evolution of high-cycle fatigue life of Ti-6Al-4V titanium alloy after multi-step hot forming is crucial for optimizing process parameters and improving service performance. The high-cycle fatigue life of Ti-6Al-4V titanium alloy after forming is influenced by various factors, such as historical loading factors, number of hot deformation cycles, deformation temperature, deformation amount, and deformation rate (process parameters), and stress amplitude, stress ratio, and loading frequency (fatigue loading parameters). An accurate high-cycle fatigue life prediction model for Ti-6Al-4V titanium alloy can help optimize multi-step hot forming parameters and also simulate and predict the high-cycle fatigue life of the titanium alloy.

[0005] In recent years, with the rapid development of computer technology and data analysis technology, artificial intelligence (AI) methods have become a widely recognized research hotspot. AI methods are increasingly being combined with traditional materials performance prediction, demonstrating significant potential in reducing trial-and-error costs and shortening R&D cycles. Exploring a machine learning-based method for predicting the high-cycle fatigue life of titanium alloys after multi-step hot deformation is of great significance for accurately predicting the high-cycle fatigue life of titanium alloys after multi-step hot deformation and improving the service reliability of titanium alloy parts. Summary of the Invention

[0006] To accurately predict the high-cycle fatigue life of Ti-6Al-4V titanium alloy after multi-step hot deformation, this invention provides a machine learning-based method for predicting the high-cycle fatigue life of Ti-6Al-4V titanium alloy after multi-step hot deformation. The main feature of this method is that it uses a machine learning model to accurately predict the high-cycle fatigue life of Ti-6Al-4V titanium alloy after multi-step hot deformation.

[0007] Traditional physical model prediction methods struggle to accurately predict the high-cycle fatigue life of Ti-6Al-4V titanium alloys after multi-step hot deformation. However, by employing machine learning methods, not only can the high-cycle fatigue life of Ti-6Al-4V titanium alloys after complex multi-step hot deformation be accurately predicted, but the modeling time can also be shortened, thus solving the problem of accurately predicting the high-cycle fatigue life of Ti-6Al-4V titanium alloys after multi-step hot deformation.

[0008] By fully leveraging the advantages of machine learning models, using multi-step thermal deformation parameters and high-cycle fatigue loading parameters as inputs and high-cycle fatigue life as output, there is no need to directly construct intermediate complex mapping relationship models, which reduces modeling time and improves prediction accuracy.

[0009] This invention is implemented as follows:

[0010] A machine learning-based method for predicting the high-cycle fatigue life of titanium alloys after multi-step hot deformation, characterized in that the method comprises:

[0011] Step 1: Cut several high-temperature tensile specimens from the diffusion-bonded Ti-6Al-4V titanium alloy material; use a UTM5504X universal electronic testing machine to conduct pre-strain and secondary tensile tests on the specimens under different temperatures, strain rates and strain values ​​to obtain tensile specimens under different multi-step hot deformation conditions.

[0012] Step 2: Cut the gauge length of the tensile specimen obtained in Step 1 to obtain a high-cycle fatigue specimen. Use a QGB-100H high-frequency fatigue testing machine to conduct high-cycle fatigue tests under different stress amplitudes and stress ratios to obtain the high-cycle fatigue life of the material under the corresponding conditions.

[0013] Step 3: Using the high-cycle fatigue life data obtained in Step 2, with thermal deformation conditions and high-cycle fatigue loading conditions as inputs and high-cycle fatigue life as output, combine the data to construct a machine learning training dataset, normalize the data to transform it to the [0,1] interval, and divide the dataset into a training set and a test set, wherein the training set uses data augmentation methods to increase the total amount of data;

[0014] Step 4: Feed the training set from Step 3 into the machine learning model, and train the model using k-fold cross-validation to finally obtain a machine learning model based on the training set learning results.

[0015] Step 5: For the machine learning model in Step 4, use the coefficient of determination. R 2 Mean absolute relative error AARE Root mean square error RMSE and mean absolute error MAE The model was established, and the trained model was used to predict the high-cycle fatigue life of titanium alloys after multi-step hot deformation.

[0016] Furthermore, in step one, the UTM 5504X universal electronic testing machine was used to conduct pre-strain and secondary tensile tests on the specimen under different strain conditions; in the multi-step high-temperature tensile test, the pre-strain and secondary deformation temperatures were 750℃, and the pre-strain and secondary deformation rates were 0.001 s⁻¹. -1 The pre-strain values ​​were 0, 0.1, and 0.2, and the secondary deformation values ​​were 0, 0.05, and 0.1. The high-cycle fatigue loading conditions included stress amplitude, stress ratio, and loading frequency. The stress amplitudes were 889 MPa, 800 MPa, 720 MPa, 648 MPa, 583 MPa, and 524 MPa, respectively. The stress ratios included 0, 0.3, and 0.5. The loading frequency was 100 Hz. A total of 60 sets of tests were conducted. The test procedure was as follows: First, the titanium alloy specimen was pre-strained and subjected to secondary deformation at a certain strain rate at the corresponding deformation temperature until the specimen was deformed to the target strain. Then, the gauge length of the specimen was machined into a high-cycle fatigue test specimen, and a high-cycle fatigue test was conducted under the specified stress amplitude, stress ratio, and loading frequency.

[0017] Furthermore, the titanium alloy sample was a Ti-6Al-4V titanium alloy that had undergone diffusion bonding after being held at 920 ℃ and 3 MPa for 1.5 h. Ti-6Al-4V titanium alloy is an α+β type titanium alloy with advantages such as high specific strength, high temperature resistance, and corrosion resistance, and is widely used in the manufacture of key aerospace structural components.

[0018] Furthermore, in the multi-step high-temperature deformation test, the gauge length of the tensile specimen is 48mm×12.8mm×2mm, and in the high-cycle fatigue test, the gauge length of the fatigue specimen is 4.67mm×2mm.

[0019] Furthermore, the normalization and determination coefficients mentioned in steps 2, 3, and 4... R 2 Mean absolute relative error AARE Root mean square error RMSE and mean absolute error MAE The calculation formula is:

[0020]

[0021]

[0022]

[0023]

[0024]

[0025] in, x max and x min These are the maximum and minimum values ​​in the sample data, respectively. x i * represents the first in the dataset. i The normalized values ​​of each sample; x i and y i For experimental values ​​and predicted values, The average of the experimental values. n This represents the total number of data points in the dataset.

[0026] Furthermore, the model uses the SVR machine learning model and optimizes it with the PSO algorithm. The training set is trained using k-fold cross-validation, and the output is a PSO-SVR model trained based on the training set.

[0027] Compared with the prior art, the present invention has the following advantages and effects:

[0028] (1) This invention studies the high-cycle fatigue life of Ti-6Al-4V titanium alloy after multi-step hot deformation following diffusion bonding. By establishing a material high-cycle fatigue life prediction model based on machine learning, it can help optimize the hot forming process parameters of titanium alloy and improve the service reliability of titanium alloy parts. It is also of great help in mastering the evolution law of high-cycle fatigue life after multi-step hot deformation of titanium alloy.

[0029] (2) The present invention can flexibly select the multi-step hot deformation parameters and high-cycle fatigue loading parameters according to the actual situation, and has universality. At the same time, the method of the present invention is not limited to Ti-6Al-4V titanium alloy after diffusion bonding, but can also be applied to other titanium alloys to predict the high-cycle fatigue life of the titanium alloy after multi-step hot deformation. Attached Figure Description

[0030] Figure 1 This is a comparison chart of the training set and test set for high-cycle fatigue life prediction in an embodiment of the present invention;

[0031] Figure 2 Different datasets R in the embodiments of the present invention 2 index;

[0032] Figure 3The AARE metrics for different datasets in this embodiment of the invention;

[0033] Figure 4 These are the RMSE metrics for different datasets in the embodiments of this invention;

[0034] Figure 5 The MAE indexes for different datasets in this embodiment of the invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the following examples provide a more detailed description of the invention. It should be noted that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the invention.

[0036] The method of the present invention includes the following steps:

[0037] Step 1: Cut several high-temperature tensile specimens from the diffusion-bonded Ti-6Al-4V titanium alloy material; use a UTM5504X universal electronic testing machine to conduct pre-strain and secondary tensile tests on the specimens under different strain conditions to obtain tensile specimens under different multi-step hot deformation conditions.

[0038] Step 2: Using the gauge length of the tensile specimen obtained in Step 1, high-cycle fatigue specimens are obtained by cutting. High-cycle fatigue tests are conducted under different stress amplitudes and stress ratios using a QGB-100H high-frequency fatigue testing machine to obtain the high-cycle fatigue life of the material under the corresponding conditions. For the multi-step high-temperature deformation test, the gauge length of the tensile specimen is 48mm × 12.8mm × 2mm, and for the high-cycle fatigue test, the gauge length of the fatigue specimen is 4.67mm × 2mm.

[0039] In the multi-step high-temperature tensile test, the pre-strain and secondary strain temperatures were 750℃, and the pre-strain and secondary deformation rates were 0.001 s⁻¹. -1 The pre-strain values ​​were 0, 0.1, and 0.2, and the secondary deformation values ​​were 0, 0.05, and 0.1. The high-cycle fatigue loading conditions included stress amplitude, stress ratio, and loading frequency. The stress amplitudes were 889 MPa, 800 MPa, 720 MPa, 648 MPa, 583 MPa, and 524 MPa, respectively. The stress ratios included 0, 0.3, and 0.5. The loading frequency was 100 Hz. A total of 60 sets of tests were conducted. The test procedure was as follows: the titanium alloy specimen was first subjected to pre-straining and secondary deformation at a certain strain rate at the corresponding deformation temperature until the specimen was deformed to the target strain. Then, the gauge length of the specimen was machined into a high-cycle fatigue test specimen, and a high-cycle fatigue test was conducted under the specified stress amplitude, stress ratio, and loading frequency.

[0040] The titanium alloy sample was a Ti-6Al-4V titanium alloy that had undergone diffusion bonding after being held at 920 ℃ and 3 MPa for 1.5 h. Ti-6Al-4V titanium alloy is an α+β type titanium alloy with advantages such as high specific strength, high temperature resistance, and corrosion resistance, and is widely used in the manufacture of key aerospace structural components.

[0041] Step 3: Using the high-cycle fatigue life data obtained in Step 2, with thermal deformation conditions and high-cycle fatigue loading conditions as inputs and high-cycle fatigue life as output, combine the data to construct a machine learning training dataset, normalize the data to transform it to the [0,1] interval, and divide the dataset into a training set and a test set, wherein the training set uses data augmentation methods to increase the total amount of data;

[0042] Step 4: Feed the training set from Step 3 into the machine learning model, and train the model using k-fold cross-validation to finally obtain a machine learning model based on the training set learning results.

[0043] Step 5: For the machine learning model in Step 4, use the coefficient of determination R. 2 The mean absolute relative error (AARE), root mean square error (RMSE), and mean absolute error (MAE) were used to establish the model. The trained model was then used to predict the high-cycle fatigue life of titanium alloys after multi-step hot deformation. The model used was the SVR machine learning model, optimized using the PSO algorithm. K-fold cross-validation was employed on the training set for training, and the resulting PSO-SVR model was output based on the training set.

[0044] The normalization and determination coefficient R mentioned in steps 2, 3, and 4 of this invention 2 The formulas for calculating the mean absolute relative error (AARE), root mean square error (RMSE), and mean absolute error (MAE) are as follows:

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] in, x max and x min These are the maximum and minimum values ​​in the sample data, respectively. x i * represents the first in the dataset. iThe normalized values ​​of each sample; x i and y i For experimental values ​​and predicted values, The average of the experimental values. n This represents the total number of data points in the dataset.

[0051] like Figure 1 As shown, the data points are mostly located near the 0% deviation line, indicating that the model has good predictive performance and no overfitting phenomenon. This shows that the model can accurately predict the high-cycle fatigue life after multi-step thermal deformation and has good generalization ability.

[0052] Figure 2 , 3 4 and 5 represent the training set, test set, and overall dataset, respectively. R 2 , AARE , RMSE and MAE The model demonstrates excellent performance metrics on the training set, test set, and overall dataset, further illustrating its good prediction accuracy.

[0053] The above process describes the high-cycle fatigue life prediction method for Ti-6Al-4V titanium alloy after diffusion bonding and multi-step hot deformation. The prediction method can also be used for different multi-step hot deformation parameters and high-cycle fatigue loading parameters, and has universality. At the same time, the method of the present invention is not limited to Ti-6Al-4V titanium alloy after diffusion bonding, but can also be applied to other titanium alloys to predict the high-cycle fatigue life of the titanium alloy after multi-step hot deformation.

[0054] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A machine learning-based method for predicting the high-cycle fatigue life of titanium alloys after multi-step hot deformation, characterized in that, The method is as follows: Step 1: Cut several high-temperature tensile specimens from the diffusion-bonded Ti-6Al-4V titanium alloy material; conduct pre-strain and secondary tensile tests on the specimens under different temperatures, strain rates and strain values ​​to obtain tensile specimens under different multi-step hot deformation conditions. Step 2: Use the gauge length of the tensile specimen obtained in Step 1 to cut a high-cycle fatigue specimen, and conduct high-cycle fatigue tests under different stress amplitudes and stress ratios to obtain the high-cycle fatigue life of the material under the corresponding conditions. Step 3: Using the high-cycle fatigue life data obtained in Step 2, with thermal deformation conditions and high-cycle fatigue loading conditions as inputs and high-cycle fatigue life as output, combine the data to construct a machine learning training dataset, normalize the data to transform it to the [0,1] interval, and divide the dataset into a training set and a test set, wherein the training set uses data augmentation methods to increase the total amount of data; Step 4: Feed the training set from Step 3 into the machine learning model, and train the model using k-fold cross-validation to finally obtain a machine learning model based on the training set learning results. Step 5: For the machine learning model in Step 4, use the coefficient of determination R. 2 The mean absolute relative error (AARE), root mean square error (RMSE), and mean absolute error (MAE) were used to establish the model. The trained model was then used to predict the high-cycle fatigue life of titanium alloys after multi-step hot deformation.

2. The method for predicting high-cycle fatigue life of titanium alloys after multi-step hot deformation based on machine learning according to claim 1, characterized in that, In step one, the UTM 5504X universal electronic testing machine is used to conduct pre-strain and secondary tensile tests on the specimen under different strain values. In the multi-step high-temperature tensile test, the pre-strain and secondary deformation temperatures were 750℃, and the pre-strain and secondary deformation rates were 0.001 s⁻¹. -1 The prestrain values ​​were 0, 0.1, and 0.2, and the secondary deformation values ​​were 0, 0.05, and 0.

1. The high-cycle fatigue loading conditions included stress amplitude, stress ratio, and loading frequency. The stress amplitudes were 889 MPa, 800 MPa, 720 MPa, 648 MPa, 583 MPa, and 524 MPa, respectively. The stress ratios included 0, 0.3, and 0.

5. The loading frequency was 100 Hz. A total of 60 sets of tests were conducted.

3. The method for predicting high-cycle fatigue life of titanium alloys after multi-step hot deformation based on machine learning according to claim 2, characterized in that, The test procedure is as follows: First, the titanium alloy specimen is pre-strained and then deformed at a certain strain rate at the corresponding deformation temperature until the specimen is deformed to the target strain. Then, the gauge length section of the specimen is machined into a high-cycle fatigue test specimen, and a high-cycle fatigue test is carried out under the specified stress amplitude, stress ratio and loading frequency.

4. The method for predicting high-cycle fatigue life of titanium alloys after multi-step hot deformation based on machine learning according to claim 3, characterized in that, The titanium alloy sample was a Ti-6Al-4V titanium alloy that had undergone diffusion bonding at 920 ℃ and 3 MPa for 1.5 h. The Ti-6Al-4V titanium alloy is an α+β type titanium alloy.

5. The method for predicting high-cycle fatigue life of titanium alloys after multi-step hot deformation based on machine learning according to claim 2, characterized in that, In the multi-step high-temperature deformation test, the gauge length of the tensile specimen is 48mm×12.8mm×2mm, and in the high-cycle fatigue test, the gauge length of the fatigue specimen is 4.67mm×2mm.

6. The method for predicting high-cycle fatigue life of titanium alloys after multi-step hot deformation based on machine learning according to claim 1, characterized in that, The normalization and coefficient of determination R mentioned in steps 2, 3, and 4 are described above. 2 The formulas for calculating the mean absolute relative error (AARE), root mean square error (RMSE), and mean absolute error (MAE) are as follows: Where, x max and x min These are the maximum and minimum values ​​in the sample data, respectively; x i * represents the normalized value of the i-th sample in the dataset; x i and y i For experimental values ​​and predicted values, is the average of the experimental values, and n is the total number of data points in the dataset.

7. The method for predicting high-cycle fatigue life of titanium alloys after multi-step hot deformation based on machine learning according to claim 1, characterized in that, The model uses the SVR machine learning model and optimizes it with the PSO algorithm. The training set is trained using k-fold cross-validation, and the output is a PSO-SVR model trained based on the training set.

Images