A data-driven method for constructing high-temperature flow stress and hot processing map of titanium alloy

Abstract

This invention proposes a data-driven method for constructing high-temperature rheological stress and hot working diagrams for titanium alloys. By constructing a titanium alloy composition-Arrhenius constitutive equation parameter dataset, developing a machine learning prediction model for these parameters, and conducting thermo-mechanical coupled finite element simulation calculations, the method achieves automatic calculation of high-temperature rheological stress and rapid generation of high-temperature stress-strain curves based solely on user input of the titanium alloy composition. Furthermore, it automates the entire process of constructing hot working diagrams. This method effectively solves the problems of long construction cycles, high costs, and heavy reliance on Gleeble hot compression simulation experimental data in traditional titanium alloy high-temperature rheological stress and hot working diagram construction. It enables efficient, rapid, and accurate prediction of hot working diagrams for titanium alloys with new compositions and systems, providing a novel method for rapid iterative development of titanium alloy materials and optimization of complex process parameters, further optimizing the titanium alloy process parameter design process.

Description

Technical Field

[0001] This invention belongs to the field of materials science and engineering technology, specifically relating to a data-driven method for constructing high-temperature rheological stress and thermal processing diagrams of titanium alloys. Background Technology

[0002] Titanium alloys are widely used in high-end manufacturing fields such as aerospace and marine engineering due to their excellent low density, high specific strength, and corrosion resistance. However, titanium alloys exhibit significant sensitivity to microstructure evolution and a narrow hot working window during high-temperature hot deformation. Improper process parameters can easily lead to microstructural anomalies, excessively large grain sizes, and adiabatic shear bands, resulting in material performance degradation or even scrapping. To accurately guide the hot working process, hot working diagrams have become an important tool for process parameter design. Hot working diagrams are typically based on Dynamic Material Model (DMM) theory, combined with the material's high-temperature rheological behavior, to plot the power dissipation rate distribution and flow stability criteria at different temperatures and strain rates, thereby determining the safe processing zone and unstable zone, providing a basis for process window optimization. Therefore, constructing accurate hot working diagrams is of great guiding significance for optimizing process parameters and avoiding hot working defects in the actual production and preparation of titanium alloys.

[0003] Traditional methods for constructing heat-working maps primarily rely on experimental measurements and physical modeling. This typically requires numerous Gleeble thermal simulations at various temperatures and strain rates to obtain high-temperature rheological stress-strain curves. Subsequently, the experimental data is fitted with Arrhenius constitutive equation parameters and the energy dissipation, calculation, and instability criterion calculations in the DMM are performed to construct the heat-working map. This process is not only costly, time-consuming, and cumbersome in data processing, but also requires the complete implementation of the above experimental and analytical steps for each specific composition or microstructure of titanium alloy, making it difficult to comprehensively cover the influence of complex compositional systems and microstructure variations on the high-temperature rheological behavior of materials. Currently, some researchers are attempting to introduce machine learning techniques into traditional methods, proposing an improved method for constructing high-temperature rheological stress and heat-working maps of titanium alloys. Based on high-temperature rheological stress data from Gleeble thermal simulations, a dataset of temperature, strain rate, true strain values, and corresponding high-temperature rheological stresses for titanium alloys with specific compositions is constructed. Regression prediction models are then established using algorithms such as artificial neural networks, enabling prediction functions with temperature, strain rate, and true strain values ​​as inputs and high-temperature rheological stress as output. While this method improves the accuracy of high-temperature rheological stress prediction to some extent, it still relies heavily on experimental data. Furthermore, the established models are often limited to specific compositions, lacking intelligence and generalization ability, and cannot predict the high-temperature rheological behavior of titanium alloys under new compositions, microstructures, or processing conditions. Therefore, although this method has its improvements, it still falls short of meeting the urgent needs of rapid iterative development and optimization of complex process parameters for titanium alloy materials. Summary of the Invention

[0004] In view of this, the present invention proposes a data-driven method for constructing high-temperature rheological stress and hot working diagrams of titanium alloys. This method can predict high-temperature rheological stress with alloy composition as input, and combine DMM and instability criteria to quickly and effectively construct hot working diagrams of titanium alloys, solving the problem that the construction of hot working diagrams in the prior art depends on experimental data and has a long cycle.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows: (Consistent with the contents of the claims) Beneficial effects

[0006] 1. This invention establishes a method for predicting high-temperature rheological stress and hot working diagrams of titanium alloys based on their composition. By extracting massive amounts of literature data to construct a titanium alloy composition-Arrhenius constitutive equation parameter dataset, and employing CatBoost (categorical boosting) and Gradient Boosting Decision Tree (GBDT) machine learning models, Multiple Grey Wolf Optimization (MGWO) algorithm, and thermo-mechanical coupled finite element simulation calculations, this invention effectively realizes the fully automated function of predicting high-temperature rheological stress and constructing hot working diagrams using only the titanium alloy composition as input, providing strong technical support for actual production and preparation processes.

[0007] 2. The method established in this invention constructs a titanium alloy composition-Arrhenius constitutive equation parameter dataset. Using an open-source Large Language Model (LLM) and Prompt Engineering (PE), novel prompt words are designed to extract information such as alloy composition, phase transformation point, test temperature, true strain value, and Arrhenius constitutive equation parameters from literature. Furthermore, Python code is designed to batch extract data from massive amounts of literature, enabling rapid and efficient construction of the required dataset.

[0008] 3. The method established in this invention constructs a machine learning model for titanium alloy composition-Arrhenius constitutive equation parameters. This model only requires the user to input the titanium alloy composition, automatically calculates its molybdenum and aluminum equivalents, and divides it into β-phase and α+β-phase regions according to the test temperature and phase transformation point temperature. It outputs the corresponding Arrhenius equation constitutive parameters under different phase regions and different true strain values. Based on the constitutive parameter prediction results, the high-temperature rheological stress is automatically calculated and visualized, facilitating subsequent retrieval of relevant information.

[0009] 4. A finite element thermo-mechanical coupling calculation model is constructed in the method established in this invention. This model uses high-temperature rheological stress as the material property parameter to perform Gleeble hot compression finite element simulation calculations under different temperatures and strain rates, and extracts the simulation results in batches. Based on the finite element calculation results, combined with the DMM and instability criteria, a Python program is written to quickly construct the thermal processing diagram. Attached Figure Description

[0010] Figure 1 These are the prompt words needed to extract the parameters of the Arrhenius constitutive equation from the literature in this invention.

[0011] Figure 2This is the distribution of the Arrhenius constitutive equation parameter dataset for titanium alloy composition in this invention.

[0012] Figure 3 This is a schematic diagram of the MGWO hyperparameter optimization algorithm in this invention.

[0013] Figure 4 The high-temperature rheological stress curve of the Ti-5Mo-5Al-5V-1Cr-1Fe titanium alloy in Example 1 of this invention.

[0014] Figure 5 This is the high-temperature rheological stress curve of TB2 titanium alloy in Example 2 of the present invention.

[0015] Figure 6 This is a schematic diagram of the finite element batch simulation calculation in Embodiment 2 of the present invention.

[0016] Figure 7 This is a hot working diagram of TB2 titanium alloy under a strain value of 0.3 in Example 2 of the present invention. Detailed Implementation

[0017] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] This invention provides a data-driven method for constructing high-temperature rheological stress and hot working diagrams for titanium alloys. By extracting titanium alloy composition-Arrhenius constitutive equation parameter data from a vast amount of literature and constructing a machine learning model based on ensemble learning models Catboost, GBDT, and MGWO hyperparameter optimization algorithms, it achieves high-temperature rheological stress prediction using only titanium alloy composition as input. Combined with finite element simulation calculations, DMM, and instability criteria, it can quickly and effectively construct hot working diagrams for titanium alloys, providing a new method for designing novel hot working process parameters for titanium alloys. This method significantly shortens the design cycle while reducing the cost of existing methods, solving the problem that the construction of hot working diagrams in existing technologies relies on experimental data and has a long cycle. This invention includes the following steps: Step 1, Data Extraction and Dataset Construction: Using keywords such as "Arrhenius, titanium," "Hotdeformation behavior of titanium," "Processing map titanium," and "Titanium alloy hot working diagram" from Elsevier, CNKI, and other literature databases, 841 high-quality articles highly relevant to titanium alloy hot working diagrams were retrieved. Further analysis was conducted using designed prompts (such as...). Figure 1As shown in the figure, the system calls the AI ​​Large Language Model API interface to quickly extract the required key information from the literature. Based on the extracted information, Python code is written for data cleaning and processing, standardizing the alloy composition format, and extracting the specific element content. The corresponding molybdenum and aluminum equivalents are calculated based on the element content. The microstructure type is determined based on the alloy phase transformation point in the extracted information (if none is found, the commercial software JMAPT is used for calculation) and the Gleeble test temperature. Alloys with test temperatures above the phase transformation point are in the β phase region and are marked as 1; otherwise, they are in the α+β phase region and are marked as 2. If the true strain value is not explicitly given in the literature, the true strain value corresponding to the heat treatment diagram in the literature is used. If the heat treatment diagram is also not given, the default true strain value is 0.6. If the true strain value is expressed as the strain value at the peak stress in the literature, it is uniformly set to 0.05. The final dataset includes 23 elements and their corresponding molybdenum equivalent, aluminum equivalent, microstructure, true strain value, and Arrhenius constitutive equation parameters Q, n, LNA, and alpha. The data distribution is as follows: Figure 2 As shown.

[0019] Step 2: Establish a machine learning model for the titanium alloy composition-Arrhenius constitutive equation parameters: Using the elemental composition, molybdenum equivalent, aluminum equivalent, microstructure, and true strain values ​​in the dataset as inputs, and the Arrhenius constitutive equation parameters Q, n, LNA, and alpha as outputs, a comparative analysis was conducted to select the Catboost model and the GBDT model in the ensemble learning approach. The Catboost model was chosen for equation parameters Q and LNA, while the GBDT model was chosen for equation parameters n and alpha. The parameter n considers the constraints of the physical equations, taking not only the aforementioned elemental composition, molybdenum equivalent, aluminum equivalent, microstructure, and true strain values ​​as inputs, but also the Arrhenius constitutive equation parameters Q, LNA, and alpha as inputs. The MGWO optimization algorithm was used to optimize the hyperparameters of the machine learning model; the specific process is as follows... Figure 3 As shown. The final machine learning model prediction accuracy (test set R) 2 The performance of the four parameters Q, n, LNA, and alpha are 0.882, 0.811, 0.873, and 0.905, respectively. The model provides multiple options for the test temperature range, strain rate range, and true strain value; users can use the model defaults or customize relevant parameters. High-temperature rheological stress of the alloy under different temperatures, strain rates, and strain conditions is calculated based on the Arrhenius constitutive parameters predicted by the machine learning model.

[0020] Step 3, Construction and simulation calculation of the thermo-coupling finite element model: A 1 / 4-inch hot-compression finite element model was constructed using lsprepost software, and finite element simulation calculations were performed using Ansys_LSDYNA software. The high-temperature rheological stress obtained from the predicted model was used as input. Python code was used to batch modify k-files and generate batch files for performing batch finite element calculations under different temperatures and strain rates. The simulation results were then extracted by calling lsprepost software in a batch processing manner. Based on the simulation results, the calculation results of the Arrhenius constitutive equation were further corrected, coupling the dual influencing factors of the material constitutive equation and the actual sample structure to accurately predict the high-temperature rheological stress of the alloy.

[0021] Step 4, Construction of the thermal processing diagram: Based on the above calculation results, an alloy hot working diagram was constructed by combining the DMM and Prasad instability criteria. The entire process was implemented using Python code. The DMM treats the workpiece during hot deformation as a nonlinear power dissipator. The total input power (P) is decomposed into two parts: the heat dissipation for material plastic deformation (G content) and the dissipation co-content for microstructure changes (J co-content).

[0022] Where G represents heat dissipation due to plastic deformation, and J represents heat dissipation due to microstructure evolution, such as dynamic recovery and dynamic recrystallization. The strain rate sensitivity factor m is defined as:

[0023] When the true strain value and temperature are constant, the rheological stress and strain rate have the following relationship:

[0024] Therefore, the microstructure evolution dissipation J can be calculated by the following formula:

[0025] The maximum value of the strain rate sensitivity factor is 1, at which point J reaches its maximum value. max :

[0026] According to J and J max The energy dissipation rate η is defined as follows:

[0027] Therefore, the energy dissipation rate η can be calculated based on the strain rate sensitivity factor m. Regions with high η values ​​typically correspond to a "safe" processing window conducive to the formation of a uniform, fine-grained microstructure. Furthermore, it should be noted that during hot deformation, excessively low temperatures or excessively high strain rates can easily lead to deformation instability. Prasad et al., based on the theory of continuous large plastic deformation, proposed a criterion for deformation stability:

[0028] When the deformation instability criterion ξ is less than 0, deformation instability conditions such as local shearing, adiabatic heating, and the initiation of pores or cracks are prone to occur. Avoiding these unstable regions is key to preventing severe microstructure inhomogeneity and defects. Overlaying the contour map of the energy dissipation rate η with the unstable regions creates a hot working diagram.

[0029] Example 1: High-Temperature Rheological Stress Prediction of Ti-5Mo-5Al-5V-1Cr-1Fe Titanium Alloy The Ti-5Mo-5Al-5V-1Cr-1Fe titanium alloy was subjected to a temperature range of 700℃~850℃ and a 0.01s s⁻¹. -1 ~10s -1 The rheological stress was predicted under the following conditions. First, the alloy composition was manually input into the prediction model according to the format "Ti-5Mo-5Al-5V-1Cr-1Fe". The model automatically calculated the molybdenum equivalent to 7.1, the aluminum equivalent to 5.0, and the phase transformation temperature to 875℃. Based on the test temperature and the phase transformation temperature, the model selected a microstructure of 2 (α+β phase region), and the true strain values ​​were the model's default values ​​of 0.1, 0.2, 0.3, 0.4, 0.5, and 0.6. The model then used a machine learning module to predict the Arrhenius constitutive equation parameters under different true strain values. Based on the parameter prediction results, the calculated temperature conditions were 700, 750, 800, and 850℃, and the strain rate condition was 0.01 s⁻¹. -1 0.1s -1 1s -1 0.01s -1 The rheological stress is visualized, and the high-temperature rheological stress curve is shown below. Figure 4 As shown.

[0030] Example 2: Prediction of High-Temperature Rheological Stress and Hot Working Diagram of TB2 Titanium Alloy For TB2 (Ti-5Mo-5V-8Cr-3Al) titanium alloy at 800℃~950℃, 0.001 s -1 ~1 s -1The rheological stress and hot working curves under the specified conditions were used for prediction. First, the alloy composition was manually input into the prediction model according to the format "Ti-5Mo-5V-8Cr-3Al". The model automatically calculated its molybdenum equivalent to 15.35, aluminum equivalent to 3.0, and phase transformation temperature to 687℃. Based on the test temperature and phase transformation temperature, the model selected a microstructure of 1 (β phase region), and the true strain values ​​were set to the model's default values ​​of 0.1, 0.2, 0.3, 0.4, 0.5, and 0.6. The prediction process was the same as in Example 1, and the predicted high-temperature rheological stress curves were as follows: Figure 5 As shown. Using this result as input for the material's mechanical properties, k-files under different temperatures and strain rates are generated in batches using Python code. Python code is used to generate calculation batch files and result extraction batch files to perform batch finite element simulation calculations and extract simulation results, such as... Figure 6 As shown. Based on the simulation results, combined with the DMM model and Prasad instability criterion, the hot working diagram of TB2 titanium alloy under the condition of true strain of 0.3 is obtained as follows. Figure 7 As shown, region I is the deformation instability zone, typically containing adiabatic shear bands and inhomogeneous deformation structures. Region II is the incomplete dynamic recrystallization zone. In this region, the grain structure is inhomogeneous, with some areas still retaining deformed structures and original grain boundaries not completely eliminated, forming a mixed structure where deformed and recrystallized grains coexist. Region III is the relatively complete dynamic recrystallization zone, and also the optimal deformation zone, exhibiting a high η≥0.54.

[0031] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data-driven method for constructing high-temperature rheological stress and hot working diagrams of titanium alloys, characterized in that, Includes the following steps: Step 1: Establish a dataset of titanium alloy composition-Arrhenius constitutive equation parameters.

2. Step 2: Construction of a machine learning model for titanium alloy composition-Arrhenius constitutive equation parameters. The Arrhenius constitutive equation parameters under different strain values ​​are predicted by inputting only the titanium alloy composition information, and the high-temperature rheological stress under different temperatures and strain rates is automatically calculated.

3. Step 3: Thermo-coupling finite element model construction and simulation calculation. The high-temperature rheological stress obtained in Step 2 is used as the material property input to carry out batch hot compression finite element simulation calculation. The material constitutive equation and the actual sample structure are coupled to correct the high-temperature rheological stress of the material.

4. Step four: Construction of hot working diagrams. Based on the high-temperature rheological stress data corrected in step three, and combined with the dynamic material model and instability criteria, hot working diagrams under different strain conditions are automatically constructed.

5. The method as described in claim 1, characterized in that, In step one, the data entries in the dataset are derived from 841 high-quality articles that are highly relevant to the hot working diagrams of titanium alloys. Data is extracted in batches by designing prompt words and calling an artificial intelligence large language model.

6. The method as described in claim 1, characterized in that, In step two, the Arrhenius constitutive equation parameters Q and LNA are optimized using the Catboost machine learning model and the hybrid gray wolf optimization algorithm; the Arrhenius constitutive equation parameters n and alpha are optimized using the GBDT machine learning model and the hybrid gray wolf optimization algorithm. The inputs to the prediction models for parameters Q, LNA, and alpha are alloying elements and their contents, molybdenum equivalent, aluminum equivalent, microstructure, and true strain value. In addition to the above inputs, parameter n also receives inputs from parameters Q, LNA, and alpha.

7. The method as described in claim 1, characterized in that, In step three, based on the high-temperature rheological stress data under different conditions in step two, Python code is used to programmatically modify the temperature and material keywords in the k-file in batches, generate batch calculation files, and generate batch result extraction files, so as to obtain the corrected high-temperature rheological stress under the dual factors of coupled material constitutive equation and actual sample structure.

8. The method as described in claim 1, characterized in that, In step four, based on the high-temperature rheological stress obtained in step three, the thermal processing diagram is drawn programmatically using Python code.

9. The method as described in claim 2, characterized in that, In step one, the dataset covers information on 23 alloying elements, and also includes information on molybdenum equivalent, aluminum equivalent, microstructure, true strain value, Arrhenius constitutive equation parameter Q, Arrhenius constitutive equation parameter n, Arrhenius constitutive equation parameter LNA, and Arrhenius constitutive equation parameter alpha. There are 774 data entries.

10. The method as described in claim 3, characterized in that, In step two, the user only needs to input the titanium alloy composition information, and the model automatically calculates the molybdenum equivalent, aluminum equivalent, and phase transformation point, and determines the microstructure corresponding to the test temperature based on the phase transformation point temperature. The model provides multiple options for the test temperature range, strain rate range, and true strain value; users can use the model defaults or customize the relevant parameters.

Images