Method for designing high-temperature high-toughness titanium alloy based on machine learning and preparation method

Abstract

The invention relates to the technical field of computer aided design, in particular to a method for designing a high-temperature and high-toughness titanium alloy based on machine learning and a preparation method. The method comprises the following steps: acquiring components, a heat treatment process and corresponding tensile property data of a titanium alloy, constructing a titanium alloy data set and defining an exploration space; cleaning the titanium alloy data set, dividing the titanium alloy data set into a training set and a verification set, and normalizing the training set and the verification set; training the titanium alloy tensile property prediction model through a machine learning model and a training set, and evaluating the model through a cross validation method and a validation set; the components and heat treatment process parameters of the high-temperature and high-toughness titanium alloy are obtained through the trained prediction model, the genetic algorithm and the exploration space; according to the components of the high-temperature and high-toughness titanium alloy, the high-temperature and high-toughness titanium alloy is prepared through electric arc melting, and heat treatment is conducted according to heat treatment process parameters of the high-temperature and high-toughness titanium alloy. According to the method, collaborative intelligent design of titanium alloy components and the process is achieved, and the development efficiency is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design technology, specifically to a method and preparation method for designing high-temperature, high-strength, and high-toughness titanium alloys based on machine learning. Background Technology

[0002] Titanium alloys, as important lightweight and high-strength structural materials, have broad application prospects in aerospace, energy equipment and other fields. With the continuous improvement of the requirements for the high-temperature resistance of materials in related technical fields, the development of new titanium alloys that can maintain high strength and good toughness at temperatures of 600 degrees Celsius and above has become an important research direction. Traditionally, the composition design and process optimization of titanium alloys mainly rely on the professional experience of researchers, and the material formulation and heat treatment system are gradually improved through systematic experimental exploration. This process usually involves a large number of experiments and requires a considerable investment of time and resources.

[0003] Traditional high-temperature titanium alloys based on the "Ti-Al-Sn-Zr-Mo-Si" system still have potential for improvement in solid solution strengthening and age-hardening effects through compositional optimization. For example, in invention application CN102329983A, Baoshan Iron & Steel Co., Ltd. developed a high-temperature titanium alloy with the following weight percentage composition: Al: 5.5%~7.0%, Sn: 2.5%~4.0%, Zr: 1.0%~3.0%, Mo: 1.0%~3.0%, Nb: 1.5%~3.0%, Si: 0.10%~0.4%, Ce: 0.05%~1.0%, Ta: 0.1%~5.0%, C: 0.1%~2.0%, B: 0.1%~2.0%, with the balance being Ti and other unavoidable impurities. This alloy can be obtained through heat treatment. It exhibits high high-temperature strength, with a tensile strength reaching 772 MPa at 600°C, but its elongation is only 4.5%. In invention application CN117127059A, the Shenyang Foundry Research Institute of China National Machinery Industry Corporation developed a cast high-temperature titanium alloy. This alloy is composed of the following weight percentages: Al 4.5~5.0%, Sn 3.5~4.0%, Zr 1.8~2.2%, Mo 0.5~3.0%, Si 0.2~0.4%, Nb 0.35~1.00%, Ta 0.35~1.00%, W 0.35~1.00%, Y 0~0.04%, B 0.02~0.05%, C 0.02~0.04%, with the balance being Ti and other unavoidable impurities. At 600°C, its elongation can reach 15.5%, but its tensile strength drops to 610 MPa.

[0004] In recent years, computational methods have been increasingly applied in materials science, providing new avenues for materials research and development. In particular, machine learning technology can uncover the complex relationships between composition, processing, and properties from existing experimental data, providing a powerful tool for predicting material properties. Against this backdrop, exploring how to systematically apply data-driven methods to the design and preparation of high-temperature, high-strength, and high-toughness titanium alloys has significant research value and engineering implications. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a method for designing and preparing high-temperature, high-strength, and high-toughness titanium alloys based on machine learning.

[0006] The technical solution of this invention: a method for designing high-temperature, high-strength, and high-toughness titanium alloys based on machine learning, comprising the following specific implementation steps: S1. Obtain the composition, heat treatment process parameters, and corresponding tensile property data of titanium alloys, construct a titanium alloy dataset, and define the titanium alloy composition exploration space and heat treatment process parameter exploration space. S2. Clean the titanium alloy dataset, calculate empirical parameters as new features, divide the dataset into training and validation sets, and normalize all features. S3. Based on feature importance analysis, key features are selected, several machine learning algorithms are used for training, and the model performance is evaluated through cross-validation and validation set. The stacked ensemble model is selected as the prediction model for the tensile properties of titanium alloys. S4. Embed the trained stacked ensemble model into the genetic algorithm, set the tensile test temperature and heat treatment constraints, perform iterative search in the exploration space, and output the composition and heat treatment process parameter combination of the high-temperature high-strength and high-toughness titanium alloy. S5. Based on the output composition, high-purity raw materials are used for arc melting, and heat treatment is carried out according to the output heat treatment process parameters to prepare high-temperature, high-strength, and high-toughness titanium alloy.

[0007] Preferably, the titanium alloy dataset construction process is as follows: Data on the composition, processing technology, and tensile properties of titanium alloys were retrieved from literature, patents, and Matweb databases, and compiled into a dataset suitable for modeling. The dataset samples are characterized by titanium alloy composition, heat treatment parameters, and tensile test temperature, and are labeled with tensile strength and elongation at break.

[0008] Preferably, the titanium alloy composition is: Al: 5-7%, Sn: 0.5-5.0%, Zr: 2-6%, Nb: 1-4%, Mo: 1-5%, Si: 0.1-0.8%, Ta: 0.5-4.0%, W: 0.1-5.0%, V: 0-2%, Cr: 0.2-5.0%, with the balance being Ti; The exploration space for titanium alloy heat treatment process parameters is as follows: first heat treatment temperature: 800-1200℃, first heat treatment holding time: 0.5-36h, second heat treatment temperature: 500-1050℃, second heat treatment holding time: 1-150h.

[0009] Preferably, step S2 specifically includes: The dataset is cleaned, including removing noise, handling missing and outlier values, to ensure data accuracy. Calculate the empirical parameters of the samples in the dataset and use them as new features for the samples; Select relatively independent samples from the dataset as the validation set, and use the rest as the training set; The dataset is normalized to the (0,1) interval using a deviation normalization algorithm to make data from different dimensions comparable.

[0010] Preferably, empirical parameters include molybdenum equivalent, aluminum equivalent, and theoretical β-transformation temperature; The formula for calculating molybdenum equivalent is: ; Among them, Al wt Indicates the mass percentage of aluminum; Sn wt Zr represents the mass percentage of tin. wt Indicates the mass percentage of zirconium; Nb wt Indicates the mass percentage of niobium; Mo wt Indicates the mass percentage of molybdenum; Si wt Indicates the mass percentage of silicon; Ta wt Indicates the mass percentage of tantalum; W wt Indicates the mass percentage of tungsten; Fe wt V represents the mass percentage of iron. wt Indicates the mass percentage of vanadium; Cr wt This indicates the mass percentage of chromium. The formula for calculating aluminum equivalent is: ; The theoretical β transition temperature was obtained by looking up a table.

[0011] Preferably, the stacked ensemble model is constructed using a random forest regressor, an XGBoost regressor, and a gradient boosting regressor as base learners, and an elastic network regression model as a meta-learner.

[0012] Preferably, the training and evaluation process for the titanium alloy tensile property prediction model is as follows: The training set is used to train the XGboost tree model to obtain the importance of each feature of the sample. Finally, 17 key features are retained for each sample in the training set and validation set. Several machine learning algorithms were selected for model training, including random forest, XGBoost, gradient boosting regression, decision tree, support vector machine, and stacked ensemble model. All models take the preserved features as input and output tensile strength and elongation at break. Support vector machine models that do not support direct multiple outputs use the RegressorChain function from the sklearn library to pack them to achieve multiple outputs. The prediction accuracy and stability of different models were evaluated by using 5-fold cross-validation on the training set and independent validation sets. The values ​​of R2, MAE and RMSE of the prediction results of each model were obtained. The performance of each model on the above evaluation indicators was compared. Finally, the stacked ensemble model with the best balance of performance on the two labels of tensile strength and fracture elongation was selected to predict the tensile strength and fracture elongation of titanium alloy materials.

[0013] Preferably, the constraints of the genetic algorithm include a tensile test temperature of 600°C and must include two heat treatment steps.

[0014] Preferably, step S5 includes: Based on the composition of the high-temperature high-strength and high-toughness titanium alloy, the high-temperature high-strength and high-toughness titanium alloy is prepared by electric arc melting using raw materials with a purity of ≥99.9% and by repeated melting 12 times. During the electric arc melting process, high-purity argon gas is used for protection, the melting current is 300A±5A, and the melting time for each melting is ≥15 minutes. According to the heat treatment parameters of the high-temperature, high-strength and high-toughness titanium alloy, homogenization heat treatment is performed, followed by hot rolling and hot working, and finally aging heat treatment. The process parameters for hot rolling are as follows: hot rolling temperature: 950~1000℃, holding time: 0.5h, and hot rolling deformation: 68%.

[0015] The technical solution of the present invention is a method for preparing a high-temperature, high-strength, and high-toughness titanium alloy. The titanium alloy composition and heat treatment process parameters are designed using the above-mentioned machine learning-based design method for high-temperature, high-strength, and high-toughness titanium alloys, and the alloy is prepared by arc melting and heat treatment.

[0016] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention presents a method and preparation method for designing high-temperature, high-strength, and high-toughness titanium alloys based on machine learning. By integrating machine learning and genetic algorithms, it achieves the collaborative design and optimization of titanium alloy composition and heat treatment processes. By constructing a high-quality dataset and performing feature engineering, combined with the high-precision prediction capability of the stacked ensemble model, it is possible to reliably establish the complex nonlinear mapping relationship between composition, process, and performance. Furthermore, by embedding the performance prediction model into the optimization algorithm for high-throughput iterative search, it can intelligently and automatically find the optimal combination of composition and process with both high tensile strength and good elongation within a broad exploration space, thus realizing the customized design of material properties. This invention not only shortens the research and development cycle and reduces research and development costs, but also provides a reliable technical approach for developing novel titanium alloys with excellent high-temperature performance. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the process for designing high-temperature, high-strength, and high-toughness titanium alloys based on machine learning, as proposed in this invention. Detailed Implementation

[0018] Example 1, as Figure 1 As shown, the present invention proposes a method for designing high-temperature, high-strength, and high-toughness titanium alloys based on machine learning, which includes the following specific implementation steps: S1. Systematically collect data on titanium alloy composition, heat treatment processes, and tensile properties from literature, patents, and databases. Construct a dataset containing 660 samples and define the exploration space for titanium alloy composition and heat treatment process parameters. This provides a data foundation and search boundary for subsequent modeling and optimization. Specifically: The composition, heat treatment parameters, and corresponding tensile property data (including tensile strength, yield strength, and elongation at break) of titanium alloys were obtained from existing literature, patents, and online databases (such as Matweb and Matminer). The collected data included, but was not limited to, titanium alloy grades (such as TC4 and TC11), alloy element contents (such as Al, Sn, Zr, Mo, Nb, Si, Ta, W, V, and Cr), impurity element contents (such as Fe, O, N, and C), heat treatment parameters (such as temperature and holding time), and tensile test temperatures. Finally, a titanium alloy dataset containing 660 samples was constructed. Simultaneously, the exploration space for titanium alloy composition and the exploration space for heat treatment process parameters are defined: Exploring the composition of titanium alloys (by mass percentage): Al: 5-7%; Sn: 0.5-5.0%; Zr: 2-6%; Nb: 1-4%; Mo: 1-5%; Si: 0.1-0.8%; Ta: 0.5-4.0%; W: 0.1-5.0%; V: 0-2%; Cr: 0.2-5.0%; balance is Ti and other unavoidable impurities; Exploration of heat treatment process parameters: First heat treatment temperature: 800-1200℃; First heat treatment holding time: 0.5-36 hours; Second heat treatment temperature: 500-1050℃; Second heat treatment holding time: 1-150h; Each sample in the dataset is characterized by its titanium alloy composition, heat treatment parameters, and tensile test temperature, and is labeled with tensile strength and elongation at break.

[0019] S2. Clean the dataset to remove noise and outliers, calculate empirical parameters as new features, divide the dataset into training and validation sets, and use deviation standardization to normalize all features to the (0,1) interval to ensure data quality and feature comparability, preparing for model training. Specifically: Data cleaning is performed on the constructed dataset, including removing noisy data, handling missing values ​​(through deletion or imputation strategies) and outliers (identified and corrected based on statistical methods) to ensure the accuracy and consistency of the data; Calculate the molybdenum equivalent (Mo) for each sample in the dataset. eq ), aluminum equivalent (Al) eq Empirical parameters such as the theoretical β-transformation temperature were added as new features to the dataset. The formula for calculating molybdenum equivalent is: ; Among them, Al wt Indicates the mass percentage of aluminum; Sn wt Zr represents the mass percentage of tin. wt Indicates the mass percentage of zirconium; Nb wt Indicates the mass percentage of niobium; Mo wt Indicates the mass percentage of molybdenum; Si wt Indicates the mass percentage of silicon; Ta wt Indicates the mass percentage of tantalum; W wt Indicates the mass percentage of tungsten; Fe wt V represents the mass percentage of iron. wt Indicates the mass percentage of vanadium; Cr wt This indicates the mass percentage of chromium. The formula for calculating aluminum equivalent is: ; The theoretical β-transition temperature was obtained by looking up a table. Fourteen samples from independent literature were selected from the dataset as the validation set, and the remaining samples were used as the training set. The deviation normalization algorithm was then used to normalize all features in both the training and validation sets to the (0,1) interval, as shown in the formula: ; Among them, y i x represents the standardized feature value of the i-th sample obtained after calculation; i Let X represent the original feature value of the i-th sample in the dataset; min(X) represents the minimum value among all values ​​of the feature column (i.e., X) in the entire dataset; max(X) represents the maximum value among all values ​​of the feature column (i.e., X) in the entire dataset.

[0020] S3. Based on feature importance analysis, 17 key features were selected. Several machine learning algorithms were used for training. The model performance was evaluated through 5-fold cross-validation and independent validation sets. Finally, the stacked ensemble model, which showed the best and most balanced performance in predicting tensile strength and elongation at break, was selected as the tensile performance prediction model. Specifically: The training set was trained using tree models such as XGBoost, the importance of each feature was calculated, and low-importance features were removed. Finally, 17 key features were retained for each sample (Mo equivalent, Al equivalent, Tβ, Tβ-heat treatment 1 temperature, test temperature, Ywt, Erwt', Bwt, Hfwt, heat treatment 1 temperature, heat treatment 1 time, heat treatment 1 cooling method, heat treatment 2 temperature, heat treatment 2 time, heat treatment 2 cooling method, heat treatment 3 temperature, heat treatment 3 time). Multiple machine learning algorithms were selected for model training, including Random Forest, XGBoost, Gradient Boosting Regression, Decision Tree, Support Vector Machine, and Stacking Ensemble Model. The Stacking Ensemble Model was constructed by using Random Forest regressors, XGBoost regressors, and Gradient Boosting regressors as base learners and Elastic Net Model as a meta-learner. All models take the 17 retained features as input and tensile strength and elongation at break as output; for support vector machine models that do not support multiple outputs, the RegressorChain function of the sklearn library is used to pack them to achieve multiple outputs; The prediction accuracy and stability of each model were evaluated through 5-fold cross-validation on the training set and validation on the independent validation set. The evaluation metrics included R² (coefficient of determination), MAE (mean absolute error), and RMSE (root mean square error). By comparing the performance of each model on the two labels of tensile strength and elongation at break, the stacked integrated model with the most balanced performance was finally selected as the prediction model for the tensile properties of titanium alloys. The specific evaluation results are shown in Table 1 below: Table 1. Schematic diagram of evaluation results .

[0021] S4. Embed the trained stacked ensemble model into a genetic algorithm, setting a tensile test temperature of 600℃ and two-step heat treatment as constraints. By dynamically adjusting the crossover and mutation probabilities, a high-throughput iterative search is performed within the composition and heat treatment parameter exploration space to output the optimal combination of titanium alloy composition and heat treatment process parameters with the best high-temperature tensile properties (high tensile strength and good elongation), specifically: Embed the trained stacked ensemble model into the genetic algorithm and set constraints: The tensile test temperature is 600℃ and must include two heat treatment steps (i.e., a first heat treatment and a second heat treatment). The process of genetic algorithms includes population initialization, fitness evaluation (based on model-predicted tensile strength and elongation at break), selection, crossover, and mutation; By dynamically adjusting the crossover and mutation probabilities, repeated iterative searches are performed within the exploration space of titanium alloy composition and heat treatment process parameters defined in step S1 to balance global exploration and local optimization capabilities. The genetic algorithm ultimately outputs multiple combinations of composition and heat treatment process parameters for high-temperature, high-strength, and high-toughness titanium alloys. These combinations exhibit the best tensile properties (high tensile strength and good elongation) at 600℃. For example, recommended ingredients include: Ti-5.49Al-0.84Sn-3.09Zr-3.96Mo-0.64Si-3.24Nb-2.27Ta-0.65W-0.53V-2.9Cr-0.1Y; Ti-5.42Al-3.86Sn-5.06Zr-2.1Mo-0.29Si-2.14Nb-3.94Ta-4.45W-0.01V-0.24Cr-0.11Y; Ti-5.3Al-0.87Sn-3.07Zr-2.74Mo-0.11Si-3.53Nb-1.86Ta-0.18W-1.97V-0.91Cr; The corresponding heat treatment process parameters include homogenization heat treatment (e.g., water quenching after holding at 830℃ for 1.5h), hot rolling (e.g., holding at 1000℃ for 0.5h, with a deformation of 68%), and aging heat treatment (e.g., air cooling after holding at 700℃ for 8h).

[0022] S5. Based on the optimized composition, high-purity raw materials (≥99.9%) were used for arc melting. The mixture was melted 12 times at 300A under high-purity argon protection to ensure uniform composition. Subsequently, homogenization heat treatment, hot rolling (950-1000℃, holding for 0.5h, deformation 68%), and aging heat treatment were performed according to the recommended process. Finally, a titanium alloy material with high strength and toughness at 600℃ was prepared. Based on the composition recommended by the genetic algorithm, metal raw materials with a purity of ≥99.9% (such as Al, Sn, Zr, Mo, Nb, Si, Ta, W, V, Cr, Y, etc.) are used for arc melting under the protection of high-purity argon gas; the melting current is 300A±5A, the melting time for each melting is ≥15 minutes, and the melting is repeated 12 times to ensure the uniformity of composition. The smelted ingots are then processed according to the recommended heat treatment parameters: Perform homogenization heat treatment (e.g., hold at 830℃ for 1.5 hours and then water quench). The material undergoes hot rolling at a temperature of 950-1000℃, a holding time of 0.5h, and a hot rolling deformation of 68%. Perform aging heat treatment (e.g., hold at 700℃ for 8 hours and then air cool); The mechanical properties of the prepared titanium alloy samples were tested. Tensile tests were conducted at 600℃ with a strain rate of 5 × 10⁻⁶. -4 s -1 ; Tensile strength, yield strength, and total elongation are calculated using stress-strain curves; For example: The compound Ti-5.49Al-0.84Sn-3.09Zr-3.96Mo-0.64Si-3.24Nb-2.27Ta-0.65W-0.53V-2.9Cr-0.1Y has a yield strength of 731 MPa, a tensile strength of 734 MPa, and a total elongation of 11% at 600℃. The compound Ti-5.42Al-3.86Sn-5.06Zr-2.1Mo-0.29Si-2.14Nb-3.94Ta-4.45W-0.01V-0.24Cr-0.11Y has a yield strength of 500 MPa, a tensile strength of 515 MPa, and a total elongation of 32% at 600℃. The yield strength of the compound Ti-5.3Al-0.87Sn-3.07Zr-2.74Mo-0.11Si-3.53Nb-1.86Ta-0.18W-1.97V-0.91Cr at 600℃ is 118.059 MPa, the tensile strength is 148.95 MPa, and the total elongation is 46.66%.

[0023] Example 2: A method for preparing a high-temperature, high-strength, and high-toughness titanium alloy proposed in this invention is applied to a method for designing high-temperature, high-strength, and high-toughness titanium alloys based on machine learning proposed in Example 1. The target alloy composition is: Ti-5.49Al-0.84Sn-3.09Zr-3.96Mo-0.64Si-3.24Nb-2.27Ta-0.65W-0.53V-2.9Cr-0.1Y; Specifically, the implementation process is as follows: A1. Dataset Construction: Collect 660 sets of experimental data on titanium alloy composition, process parameters and tensile properties, divide them into 14 samples as validation set and the rest as training set, and construct an exploration space for titanium alloy composition and heat treatment process parameters. A2. Model Training: A stacked ensemble model is constructed using a random forest regressor, an XGBoost regressor, and a gradient boosting regressor as base learners, and an elastic network regression model as a meta learner. The model is trained using the training set in step A1 to optimize the tensile strength + elongation at break prediction model. A3. Composition Design: The model trained in step A2 is embedded into a genetic algorithm. Constraints are set (tensile test temperature is 600℃, with two heat treatment steps). The crossover and mutation probabilities are dynamically adjusted. The composition and heat treatment process parameter space in step A1 are iteratively searched. The final recommended composition is Ti-5.49Al-0.84Sn-3.09Zr-3.96Mo-0.64Si-3.24Nb-2.27Ta-0.65W-0.53V-2.9Cr-0.1Y. The recommended homogenization heat treatment process parameters are holding at 830℃ for 1.5h followed by water quenching, and the recommended aging heat treatment process parameters are holding at 700℃ for 8h followed by air cooling. The predicted tensile strength of this scheme at 600℃ is 680.96MPa, and the predicted total elongation is 21.85%. A4. Arc melting: The components recommended in step A3 are melted using metal raw materials with a purity of ≥99.9% under the protection of high-purity argon gas. The melting current is 300A, the melting time for a single melting is ≥15 minutes, and the melting is repeated 12 times to ensure the uniformity of the components. A5. Heat treatment and hot working: First, perform homogenization heat treatment according to the homogenization heat treatment process parameters recommended in step A3, then perform hot rolling at 1000℃ for 0.5h with a deformation of 68%, and finally perform aging heat treatment according to the aging heat treatment process parameters recommended in step A3 to obtain high-temperature high-strength and high-toughness titanium alloy.

[0024] Example 3: A method for preparing a high-temperature, high-strength, and high-toughness titanium alloy proposed in this invention is applied to a machine learning-based method for designing high-temperature, high-strength, and high-toughness titanium alloys proposed in Example 1. The target alloy composition is: Ti-5.42Al-3.86Sn-5.06Zr-2.1Mo-0.29Si-2.14Nb-3.94Ta-4.45W-0.01V-0.24Cr-0.11Y; Specifically, the implementation process is as follows: B1. Dataset Construction: Collect 660 sets of experimental data on titanium alloy composition, process parameters and tensile properties, divide them into 14 samples as validation set and the rest as training set, and construct an exploration space for titanium alloy composition and heat treatment process parameters. B2. Model Training: A stacked ensemble model is constructed using a random forest regressor, an XGBoost regressor, and a gradient boosting regressor as base learners, and an elastic network regression model as a meta learner. The model is trained using the training set in step B1 to optimize the tensile strength + elongation at break prediction model. B3 Composition Design: The model trained in step B2 is embedded into a genetic algorithm. To compare with Example 2, constraints are set (tensile test temperature is 600℃, only aging heat treatment). The crossover and mutation probabilities are dynamically adjusted, and the composition and heat treatment process parameter space in step B1 are iteratively searched. The final recommended composition is: Ti-5.42Al-3.86Sn-5.06Zr-2.1Mo-0.29Si-2.14Nb-3.94Ta-4.45W-0.01V-0.24Cr-0.11Y; The recommended aging heat treatment process parameters are: holding at 700℃ for 10 hours followed by air cooling. The predicted tensile strength of this method at 600℃ is 451.26 MPa, and the predicted total elongation is 33.61%. B4. Arc melting: The composition recommended in step B3 is melted using metal raw materials with a purity of ≥99.9% under the protection of high-purity argon gas. The melting current is 300A, the single melting time is ≥15 minutes, and the melting is repeated 12 times to ensure the uniformity of composition. B5. Heat treatment and hot working: First, hot rolling is carried out at 950℃ for 0.5h with a deformation of 68%. Finally, aging heat treatment is carried out according to the aging heat treatment process parameters recommended in step B3 to obtain high-temperature high-strength and high-toughness titanium alloy.

[0025] Example 4: A method for preparing a high-temperature, high-strength, and high-toughness titanium alloy proposed in this invention is applied to a method for designing high-temperature, high-strength, and high-toughness titanium alloys based on machine learning, as proposed in Example 1. The target alloy composition is: Ti-5.3Al-0.87Sn-3.07Zr-2.74Mo-0.11Si-3.53Nb-1.86Ta-0.18W-1.97V-0.91Cr; Specifically, the implementation process is as follows: C1. Dataset Construction: Collect 660 sets of experimental data on titanium alloy composition, process parameters and tensile properties, divide them into 14 samples as validation set and the rest as training set, and construct an exploration space for titanium alloy composition and heat treatment process parameters. C2. Model Training: A stacked ensemble model is constructed using a random forest regressor, an XGBoost regressor, and a gradient boosting regressor as base learners, and an elastic network regression model as a meta learner. The model is trained using the training set in step C1 to optimize the tensile strength + elongation at break prediction model. C3. Composition Design: The model trained in step C2 is embedded into a genetic algorithm. To compare with Example 3, constraints are set (tensile test temperature of 600℃, two-step heat treatment, no Y element). The crossover and mutation probabilities are dynamically adjusted, and the composition and heat treatment process parameter space in step C1 are iteratively searched. The final recommended composition is: Ti-5.3Al-0.87Sn-3.07Zr-2.74Mo-0.11Si-3.53Nb-1.86Ta-0.18W-1.97V-0.91Cr; The recommended aging heat treatment process parameters are: holding at 700℃ for 10 hours followed by air cooling. The predicted tensile strength of this method at 600℃ is 225.93 MPa, and the predicted total elongation is 45.54%. C4. Arc melting: The composition recommended in step C3 is melted using metal raw materials with a purity of ≥99.9% under the protection of high-purity argon gas. The melting current is 300A, the single melting time is ≥15 minutes, and the melting is repeated 12 times to ensure the uniformity of composition. C5. Heat treatment and hot working: First, perform homogenization heat treatment according to the homogenization heat treatment process parameters recommended in step C3, then perform hot rolling at 1000℃ for 0.5h with a deformation of 68%, and finally perform aging heat treatment according to the aging heat treatment process parameters recommended in step C3 to obtain a high-temperature, high-strength and high-toughness titanium alloy.

[0026] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method of designing high temperature high strength and toughness titanium alloys based on machine learning, characterized by, The specific implementation steps include: S1, obtain the composition of titanium alloy, heat treatment process parameters and corresponding tensile property data, construct a titanium alloy data set, and define the titanium alloy composition exploration space and heat treatment process parameter exploration space; S2, clean the titanium alloy data set, calculate the empirical parameters as new features, divide the data set into a training set and a validation set, and normalize all features; S3, based on feature importance analysis, filter key features, use several machine learning algorithms for training, evaluate model performance through cross-validation and validation set, and select a stacked ensemble model as a titanium alloy tensile property prediction model; S4, embed the trained stacked ensemble model into a genetic algorithm, set the tensile test temperature and heat treatment constraints, and perform iterative search in the exploration space to output the composition and heat treatment process parameter combination of high-temperature high-strength and high-toughness titanium alloy; S5, arc melting is performed using high-purity raw materials according to the output composition, and heat treatment is performed according to the output heat treatment process parameters to prepare high-temperature high-strength and high-toughness titanium alloy.

2. The method of designing high temperature high strength and toughness titanium alloys based on machine learning according to claim 1, wherein, The titanium alloy data set construction process is as follows: From literature, patents and Matweb database, the composition, processing technology and tensile property data of titanium alloy are queried and collected, and are arranged and constructed into a data set suitable for modeling; Among them, the features of the sample of the data set are titanium alloy composition and heat treatment parameters and tensile test temperature, and the labels are tensile strength and elongation at break.

3. The method of designing high temperature high strength and toughness titanium alloys based on machine learning according to claim 2, wherein, The titanium alloy composition exploration space is: Al: 5-7%, Sn: 0.5-5.0%, Zr: 2-6%, Nb: 1-4%, Mo: 1-5%, Si: 0.1-0.8%, Ta: 0.5-4.0%, W: 0.1-5.0%, V: 0-2%, Cr: 0.2-5.0%, and the balance is Ti; The titanium alloy heat treatment process parameter exploration space is: first heat treatment temperature: 800-1200℃, first heat treatment holding time: 0.5-36h, second heat treatment temperature: 500-1050℃, second heat treatment holding time: 1-150h.

4. The method of designing high temperature high strength and toughness titanium alloys based on machine learning according to claim 3, wherein, Step S2 specifically includes: Clean the data set, including removing noise, handling missing values and outliers, and ensuring data accuracy; Calculate the empirical parameters of the samples in the data set and use them as new features; Select relatively independent samples from the data set as a validation set, and the rest as a training set; Normalize the data set to the (0, 1) interval through the dispersion standardization algorithm to make data of different dimensions comparable.

5. The method of designing high temperature high strength and toughness titanium alloys based on machine learning according to claim 4, wherein, The empirical parameters include molybdenum equivalent, aluminum equivalent and theoretical beta transformation temperature; The calculation formula of molybdenum equivalent is: ； wherein, Al wt represents the mass percentage of aluminum element; Sn wt represents the mass percentage of tin element; Zr wt represents the mass percentage of zirconium element; Nb wt represents the mass percentage of niobium element; Mo wt represents the mass percentage of molybdenum element; Si wt represents the mass percentage of silicon element; Ta wt represents the mass percentage of tantalum element; W wt represents the mass percentage of tungsten element; Fe wt represents the mass percentage of iron element; V wt represents the mass percentage of vanadium element; Cr wt represents the mass percentage of chromium element; The calculation formula of aluminum equivalent is: ； The theoretical beta transformation temperature is calculated by table lookup.

6. The method of designing high temperature high strength and toughness titanium alloys based on machine learning according to claim 5, wherein, The stacked ensemble model is constructed by using random forest regressor, XGBoost regressor and gradient boosting regressor as base learners, and using elastic network regression model as meta-learner.

7. The method of designing high temperature high strength and toughness titanium alloys based on machine learning according to claim 6, wherein, The titanium alloy tensile property prediction model training and evaluation process is: Train the XGboost tree model with the training set to obtain the importance of each feature of the sample, and finally retain 17 key features for each sample in the training set and the validation set; Several machine learning algorithms are selected for model training, including random forest, XGBoost, gradient boosting regression, decision tree, support vector machine and stacked ensemble model; All models take reserved features as input and take tensile strength and elongation at break as output, and the support vector machine model without direct multi-output is packaged using the RegressorChain function of the sklearn library to realize multi-output; The prediction accuracy and stability of different models are verified and evaluated through 5-fold cross-validation of the training set and the independent validation set, and the values of R2, MAE and RMSE evaluation indexes of the prediction results of each model are obtained, the performances of each model on the above evaluation indexes are compared, and finally the stacked ensemble model with the best performance on the two labels of tensile strength and elongation at break is selected for predicting the tensile strength and elongation at break of the titanium alloy material.

8. The method of designing high temperature high strength and toughness titanium alloys based on machine learning according to claim 1, wherein, The constraint conditions of the genetic algorithm include that the tensile test temperature is 600 DEG C, and two-step heat treatment must be included.

9. The method of designing high temperature high strength and toughness titanium alloys based on machine learning according to claim 1, wherein, Step S5 comprises: According to the composition of the high-temperature high-strength and high-toughness titanium alloy, raw materials with a purity of greater than or equal to 99.9% are used for arc melting, and the high-temperature high-strength and high-toughness titanium alloy is prepared after 12 times of repeated melting. In the arc melting process, high-purity argon gas is used for protection, the melting current is 300A±5A, and the melting time is greater than or equal to 15 minutes each time. According to the heat treatment parameters of the high-temperature high-strength and high-toughness titanium alloy, homogenization heat treatment is carried out, followed by hot rolling hot working, and finally aging heat treatment is carried out. The process parameters of the hot rolling hot working are: hot rolling temperature: 950-1000 DEG C, holding time: 0.5h, hot rolling deformation: 68%.

10. A method of producing a high-temperature, high-strength and high-ductility titanium alloy, characterized by comprising: The titanium alloy composition and heat treatment process parameters are designed by the method according to any one of claims 1 to 9, and are prepared by arc melting and heat treatment.

Images