A Machine Learning-Based Method for High-Performance Multi-Component TiAl-Based Alloy Composition Design

Abstract

This invention provides a high-performance multi-element TiAl-based alloy composition design method based on machine learning. The method includes the following steps: acquiring composition and performance data of TiAl-based alloys to establish a dataset; training a model using the training set to establish an initial model; optimizing the initial model using the K-Fold Cross Validation algorithm to obtain an improved model; evaluating and optimizing the improved model using a test set to obtain the optimal model; inputting multiple preset alloy compositions into the optimal model to obtain corresponding first-prediction data of alloy performance; and determining the optimal alloy composition based on the first-prediction data of alloy performance. Compared to traditional experimental trial-and-error methods, the design method in this invention can solve the problems of long design cycles, low accuracy, and high costs associated with high-performance TiAl-based alloy composition design.

Description

Technical Field

[0001] This invention relates to the field of metal material preparation technology, and specifically to a high-performance multi-component TiAl-based alloy composition design method based on machine learning. Background Technology

[0002] TiAl-based alloys, due to their lightweight, high strength, and excellent resistance to high-temperature oxidation and creep, have become an ideal new type of lightweight high-temperature structural material, falling between nickel-based and cobalt-based high-temperature alloys and advanced ceramic materials, and possess great potential to replace traditional high-temperature alloys. However, TiAl-based alloys have several problems that limit their industrial application, such as poor room-temperature plasticity and insufficient processing performance, especially during hot deformation, where grain growth easily leads to alloy embrittlement. Furthermore, when the service temperature exceeds 800℃, TiAl-based alloys exhibit poor high-temperature oxidation and creep resistance, making it difficult to meet high-temperature service requirements. Current improvement methods to address these issues mainly include adding protective coatings, surface modification, and matrix alloying. However, protective coating technology still faces many unresolved issues, such as controlling coating adhesion, studying coating failure mechanisms, and improving coating service life. Similarly, surface modification technologies, mainly based on surface pre-oxidation, hot-dip galvanizing, and ion implantation, also suffer from a series of unresolved problems, such as the limited anti-oxidation time of pre-oxidation methods, the unclear mechanisms of some hot-dip galvanizing methods, and the relatively complex processing technology. Therefore, matrix alloying technology has undoubtedly become the most effective technical means to improve the properties of TiAl alloys.

[0003] With the rapid development of modern technology, the demand for novel TiAl-based alloys with excellent comprehensive performance is becoming increasingly urgent in fields such as aerospace and automotive manufacturing. To date, the microalloying composition design of TiAl alloys still heavily relies on traditional experimental trial-and-error methods. However, the influence of alloying elements on the properties of TiAl alloys is extremely complex. For example, V, Mo, B, Mn, Si, Ni, Cr, and RE (rare earth elements) can improve alloy plasticity; Nb, Mo, W, B, C, and N can improve alloy strength; Nb, Cr, W, Mo, Ta, and Si can improve alloy oxidation resistance; and Si, Er, Nb, Ta, and W can improve alloy creep resistance. This diversity of elemental effects makes it difficult to quickly and accurately design alloy compositions according to given actual performance requirements in engineering applications. Furthermore, these alloying elements often exhibit synergistic effects, the mechanisms of which are difficult to elucidate experimentally. While trial-and-error composition design methods can improve a specific aspect of an alloy's strength, plasticity, or high-temperature performance by adding a particular alloying element, this usually leads to the deterioration of other alloy properties, resulting in low efficiency, long development cycles, and high costs in the research and development of high-performance new materials. Summary of the Invention

[0004] The main objective of this invention is to provide a high-performance multi-element TiAl-based alloy composition design method based on machine learning. Compared with the traditional experimental trial-and-error method, this method can solve the problems of long design cycle, low accuracy, and high cost of TiAl-based alloy composition under high performance requirements.

[0005] To achieve the above objectives, this invention provides a high-performance multi-component TiAl-based alloy composition design method based on machine learning.

[0006] The high-performance multi-component TiAl-based alloy composition design method based on machine learning includes the following steps:

[0007] The composition and performance data of TiAl-based alloys were obtained, and a dataset was established; wherein the dataset is divided into a training set and a test set; the percentage of Al atoms in the composition of the TiAl-based alloys is 45-49 at.%.

[0008] The training set was used to train the model and establish an initial composition-property model for TiAl-based alloys;

[0009] The initial model was optimized using the K-Fold Cross Validation algorithm to obtain an improved model;

[0010] The improved model is evaluated and optimized using the test set to obtain the best model.

[0011] By inputting multiple sets of preset alloy compositions into the optimal model, the first predicted data of the alloy properties corresponding to them are obtained.

[0012] The optimal alloy composition is determined based on the first predicted data of the multi-combination alloy properties.

[0013] Furthermore, the training set contains multiple sets of training data, and each set of training data includes the first alloy composition and its corresponding first alloy performance data.

[0014] The test set contains multiple sets of test data, and each set of test data includes the second alloy composition and its corresponding second alloy performance data.

[0015] Furthermore, the establishment of the initial model includes the following steps:

[0016] Randomly select N alloying elements from the first alloy composition of the training set to form an initial variable subset;

[0017] The initial variable subset is assigned to set G, and used as the input subset to call the regression model algorithm for the first iteration to construct the regression model;

[0018] After the first iteration is completed, the input components that appear in the regression model are stored in a subset of variables, and the input components that do not appear in the regression model are discarded.

[0019] The N alloy elements in the first alloy composition of the training set are randomly selected again to form a new subset of variables, and then assigned to the set G for the next iteration.

[0020] The process is repeated until the mean absolute percentage error (MAPE) no longer decreases or the maximum number of iterations is reached, thus obtaining the initial model; the maximum number of iterations is 100 to 500.

[0021] Furthermore, the regression model algorithm adopts any one of support vector regression, decision tree regression, or gradient boosting regression tree algorithm;

[0022] Preferably, N takes the value of 3 to 5.

[0023] Furthermore, the evaluation and optimization of the improved model's predictive performance using the test set includes:

[0024] The second alloy composition is input into the improved model to obtain the second predicted data of alloy properties;

[0025] The determination coefficient R is calculated using the second predicted data of the alloy properties and the second alloy property data. 2 and mean absolute percentage error (MAPE);

[0026] Determination coefficient R 2 Whether it is greater than or equal to a preset threshold, and whether the mean absolute percentage error (MAPE) is less than or equal to a preset threshold:

[0027] When the coefficient of determination R 2 If the mean absolute percentage error (MAPE) is greater than or equal to the preset threshold and less than or equal to the preset threshold, the model optimization design is complete.

[0028] When the coefficient of determination R 2 <Preset threshold or mean absolute percentage error (MAPE)> Preset threshold, repeat the optimization design steps until the coefficient of determination R is reached. 2 ≥Preset threshold and mean absolute percentage error (MAPE) ≤Preset threshold.

[0029] Furthermore, it also includes:

[0030] TiAl-based alloys were prepared using the optimal alloy composition, and experimental data on alloy properties were obtained.

[0031] The optimal alloy composition and the experimental data are added to the dataset to form a new set of data.

[0032] The optimal model is then iteratively optimized.

[0033] Furthermore, TiAl-based alloys were prepared using a powder-coated hot isostatic pressing method, with the following process parameters:

[0034] The sintering temperature is 1200–1300℃, the holding pressure is 140–180MPa, the holding time is 4–6h, and the heating rate is 5–10℃ / min.

[0035] Furthermore, the amount of data in the training set accounts for 70-80% of the total amount of data in the dataset, the amount of data in the test set accounts for 20-30% of the total amount of data in the dataset, and the sum of the ratios of the training set and the test set is always 1.

[0036] Furthermore, the TiAl-based alloy composition includes Ti, Al elements, and at least one of Cr, Nb, Mo, W, V, Ta, C, N, Ni, Co, Sn, Si, and Cu elements.

[0037] Furthermore, the properties of the TiAl-based alloy include at least one of density, room temperature tensile strength, room temperature elongation, 750°C tensile strength, 750°C elongation, and high-temperature oxidation weight gain rate.

[0038] This invention, based on a genetic algorithm elitist strategy, combines random search and regression model algorithms. This ensures that the optimal solution obtained in each iteration (existing as input parameter variables) is retained in the next iteration, preventing the optimal solution from being destroyed by crossover and mutation due to multi-dimensional feature inputs, thus maximizing the stability of finding the optimal solution. Furthermore, the invention utilizes the K-Fold Cross Validation (K-CV) algorithm to further optimize the model hyperparameters, obtaining a TiAl-based alloy composition-performance improvement model. This constructs a mapping relationship between complex multi-element alloying elements and alloy mechanical properties, achieving efficient and accurate prediction of TiAl-based alloy composition under high-performance requirements. This helps solve the problems of long design cycles, low accuracy, and high costs associated with multi-element TiAl-based alloys, thereby promoting the engineering application of TiAl new materials.

[0039] The high-performance multi-component TiAl-based alloy composition design method based on machine learning in this invention has the following advantages:

[0040] (1) Compared with the traditional experimental trial-and-error method for material development, the high-performance multi-element TiAl-based alloy composition design method based on machine learning proposed in this invention can fully explore the implicit relationship between microalloying elements and alloy properties by analyzing a large amount of existing production and research data. It can establish an alloy composition-performance prediction model without fully understanding the synergistic effect of multiple alloying elements on alloy properties, and finally realize the efficient and low-cost optimization design of high-performance multi-element TiAl-based alloy composition, which greatly reduces production costs and cycle time.

[0041] (2) To address the problem of poor prediction accuracy of multi-dimensional feature parameter models for multi-element microalloying, this invention combines random search (RS) and regression model algorithms based on the elitist strategy of genetic algorithms. It randomly and repeatedly selects a subset of variables from the multi-dimensional feature input parameters for modeling, which can reduce computational power consumption, improve convergence efficiency, and retain the optimal parameters obtained in each iteration in the final composition-performance prediction model, thereby improving the accuracy and generalization ability of the prediction model. In addition, the prediction model is further optimized by combining the K-Fold Cross Validation algorithm, which can further improve the prediction accuracy. Attached Figure Description

[0042] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0043] Figure 1 The flowchart shows a high-performance multi-component TiAl-based alloy composition design method based on machine learning, as provided in the embodiments of the present invention. Detailed Implementation

[0044] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0045] With the continuous development of big data and artificial intelligence technologies, data-driven materials research and development is gradually becoming a new model that transforms the traditional trial-and-error approach. It can significantly improve the efficiency of materials research and development, quickly solve the problem of the difficulty in accurately matching the composition and performance of multi-element alloys, and promote the engineering application of new materials.

[0046] This invention develops a novel data-driven method for designing complex multi-element TiAl-based alloy compositions based on a genetic algorithm elitist strategy. This method unlocks the synergistic and complex implicit relationships between various microalloying elements and the properties of the parts, enabling accurate prediction and efficient design of TiAl-based alloy composition systems under high-performance requirements. It solves the problems of long design cycles, low efficiency, and high costs associated with traditional TiAl-based alloy new materials.

[0047] This invention first obtains composition and performance data of TiAl-based alloys through material databases, published literature, and experiments to establish a high-quality sample database. Based on this, machine learning methods are used to mine the synergistic and complex implicit relationships between various microalloying elements and the various properties of the parts, and an alloy composition-performance machine learning model is constructed. Subsequently, the preset multi-element TiAl-based alloy composition is input into the model to obtain its corresponding part performance, thereby quickly and accurately determining the optimal alloy composition.

[0048] Figure 1 A flowchart of a high-performance multi-component TiAl-based alloy composition design method based on machine learning is shown in an embodiment of the present invention.

[0049] like Figure 1 As shown, the high-performance multi-component TiAl-based alloy composition design method based on machine learning in this invention includes the following steps:

[0050] S1. Obtain the composition and performance data of TiAl-based alloys and establish a dataset.

[0051] As an embodiment of the present invention, the TiAl-based alloy composition includes Ti, Al elements and microalloying elements, wherein the microalloying elements are at least one of Cr, Nb, Mo, W, V, Ta, C, N, Ni, Co, Sn, Si and Cu.

[0052] It is worth mentioning that the TiAl-based alloy in this invention, as an intermetallic compound, has a significant impact on its overall performance due to the Al content. Considering the correlation between material microstructure and properties, the Al atomic percentage in the dataset ranges from 45% to 49 at.%.

[0053] Furthermore, since it is difficult to densify TiAl-based alloy powder by sintering, the publicly reported high-performance sample data are mainly obtained by fully densified preparation methods such as hot isostatic pressing and directional solidification. In order to improve the accuracy of the model, the alloy performance data in the dataset of this invention are also obtained by fully densified preparation methods to eliminate the influence of porosity.

[0054] As an embodiment of the present invention, the properties of TiAl-based alloys include at least one of density, room temperature tensile strength, room temperature elongation, 750°C tensile strength, 750°C elongation, and high-temperature oxidation weight gain rate.

[0055] In an embodiment of the present invention, the dataset is divided into a training set and a test set, that is, the training set and the test set are combined to form the dataset. The amount of data in the training set accounts for 70 to 80% of the total amount of data in the dataset, and the amount of data in the test set accounts for 20 to 30% of the total amount of data in the dataset. The sum of the ratios of the training set and the test set is always 1.

[0056] In an embodiment of the present invention, during the process of establishing a dataset using the obtained composition and performance data of TiAl-based alloys, the system will randomly and automatically generate a training set and a test set. Both the training set and the test set contain the composition of TiAl-based alloys and the corresponding alloy performance data. For ease of differentiation and management, the data in the training set is referred to as training data and is used for model establishment; the data in the test set is referred to as test data and is used for performance evaluation and optimization of the established model.

[0057] In embodiments of the present invention, the training set contains multiple sets of training data, and each set of training data includes a first alloy composition and its corresponding first alloy property data. The first alloy composition serves as input data, and the first alloy property serves as output data.

[0058] As some embodiments of the present invention, the first alloy composition includes at least one of Ti and its content, Al and its content, Cr and its content, Nb and its content, Mo and its content, W and its content, V and its content, Ta and its content, C and its content, N and its content, Ni and its content, Co and its content, Sn and its content, Si and its content, and Cu and its content.

[0059] As some embodiments of the present invention, the first alloy properties include at least one of density, room temperature tensile strength, room temperature elongation, 750°C tensile strength, 750°C elongation, and high-temperature oxidation weight gain rate.

[0060] In embodiments of the present invention, the test set contains multiple sets of test data, and each set of test data includes a second alloy composition and corresponding second alloy performance data. The second alloy composition serves as input data, and the second alloy performance data serves as output data.

[0061] As some embodiments of the present invention, the second alloy composition includes at least one of Ti and its content, Al and its content, Cr and its content, Nb and its content, Mo and its content, W and its content, V and its content, Ta and its content, C and its content, N and its content, Ni and its content, Co and its content, Sn and its content, Si and its content, and Cu and its content.

[0062] As some embodiments of the present invention, the second alloy properties include at least one of density, room temperature tensile strength, room temperature elongation, 750°C tensile strength, 750°C elongation, and high-temperature oxidation weight gain rate.

[0063] It is worth mentioning that the establishment of the dataset in this invention includes:

[0064] S1-1: Obtain composition and property data of TiAl-based alloys from materials databases, publicly published high-quality literature, and experimental data to establish a basic dataset;

[0065] S1-2, based on the percentage of Al atoms, the alloy element and alloy property data are classified and processed to obtain the initial dataset of TiAl-based alloy composition-property for different Al contents.

[0066] S1-3, normalize the initial dataset to obtain the final dataset.

[0067] S2, use the training set to train the model and establish an initial model of TiAl-based alloy composition-property.

[0068] In an embodiment of the present invention, a training set is used for model training, with the first alloy composition in the training set as input data and the first alloy performance data as output data.

[0069] The initial model establishment specifically includes the following steps:

[0070] S2-1, randomly select N alloy elements from the first alloy composition of the training set to form an initial variable subset G. i ;

[0071] S2-2, the initial variable subset G i The set G is assigned as the input subset, and the regression model algorithm is called for the first iteration to build the regression model.

[0072] S2-3, after completing the initial iteration, store the input components that appear in the regression model in the variable subset G. i In the regression model, input components that do not appear are discarded;

[0073] S2-4, randomly select N elements from the first alloy composition of the original training set to form a new variable subset G.i In the middle, it is reassigned to set G for the next iteration;

[0074] S2-5, iterate repeatedly until the mean absolute percentage error (MAPE) no longer decreases or the maximum number of steps is reached to obtain the initial model of TiAl-based alloy composition-property, and this initial model contains the best variables obtained in each iteration.

[0075] It is worth mentioning that the regression model algorithm in this invention is mainly one of Support Vector Regression (SVR), Decision Tree Regression, and Gradient Boosting Regression Tree (GBDT), which are suitable for small datasets.

[0076] The hyperparameters are selected according to the regression model algorithm category. The hyperparameter settings can be adjusted as needed and are not specifically limited.

[0077] In the embodiments of the present invention, N is selected in the range of 3 to 5. If it is too large, it will significantly increase the iteration time and computing power consumption of each modeling. If it is too small, it will easily increase the number of iteration steps and reduce the computing efficiency.

[0078] In an embodiment of the present invention, the maximum number of steps in step S2-5 is 100 to 500.

[0079] This invention combines random search (RS) and regression model algorithms to achieve stepwise dimensionality reduction of multi-element alloy input features through multiple iterative calculations. This solves the problems of long establishment cycle, high computing power requirements, and low prediction accuracy of TiAl-based alloy composition-performance prediction models under complex multi-dimensional feature input conditions, thereby realizing the efficient design of high-performance multi-element TiAl-based alloy compositions.

[0080] S3 uses the K-Fold Cross Validation algorithm to optimize the hyperparameters of the initial model, resulting in an improved model.

[0081] In the embodiments of the present invention, K takes the values ​​5 and 10 in the K-Fold Cross Validation algorithm.

[0082] It should be noted that the value of K can be 5 or 10, and of course, it can be designed according to actual needs without specific requirements.

[0083] S4 uses the test set to evaluate the prediction performance of the improved model and optimize its design, thereby improving the model's accuracy and generalization ability and obtaining the best model.

[0084] In the embodiments of the present invention, the evaluation index used is the coefficient of determination R. 2 Mean Absolute Percentage Error (MAPE) is expressed as follows:

[0085]

[0086] Where y is the true value; This is a predicted value; is the average of the true values; n is the sample size of the test set.

[0087] In embodiments of the present invention, model evaluation and optimization design specifically include the following steps:

[0088] S4-1, Input the second alloy composition into the improved model to obtain the second predicted data of alloy properties;

[0089] S4-2, using the second predicted data of alloy properties and the second alloy property data, the coefficient of determination R is calculated. 2 and mean absolute percentage error (MAPE);

[0090] S4-3, Determine the coefficient of determination R. 2 Whether it is greater than or equal to a preset threshold, and whether the mean absolute percentage error (MAPE) is less than or equal to a preset threshold:

[0091] When the coefficient of determination R 2 If the model hyperparameter optimization design is completed when the mean absolute percentage error (MAPE) is ≥ the preset threshold and ≤ the preset threshold, then the improved model is the optimal model.

[0092] When the coefficient of determination R 2 <Preset threshold or mean absolute percentage error (MAPE)> Set the preset threshold, reset the hyperparameters, and repeat optimization steps S3 and S4-1 to S4-2 until the coefficient of determination R... 2 If the mean absolute percentage error (MAPE) is greater than or equal to the preset threshold and less than or equal to the preset threshold, the optimization design is complete and the best model is obtained.

[0093] In an embodiment of the present invention, R 2 The preset threshold can be 0.9, and the preset threshold for MAPE can be 10%. Stricter error limits are beneficial to improving the generalization ability of the model.

[0094] In embodiments of the present invention, the model evaluation index used includes the coefficient of determination R. 2 And mean absolute percentage error (MAPE), R 2 Its value is in the range of 0 to 1, and the closer it is to 1, the higher the model accuracy. The smaller the MAPE value, the higher the model accuracy.

[0095] S5 inputs multiple preset alloy compositions into the optimal model to obtain the first predicted data of the alloy properties corresponding to them.

[0096] In embodiments of the present invention, multiple alloy compositions can be preset and input into the TiAl-based alloy composition-performance optimal model to obtain the corresponding first predicted data of the multiple alloy performance.

[0097] S6 determines the optimal alloy composition based on the first prediction data of multiple alloy properties.

[0098] In the embodiments of the present invention, the optimal alloy composition can be screened and determined by comparing and analyzing the required alloy performance with the first predicted alloy performance data, so as to achieve accurate prediction and efficient design of TiAl-based alloy composition system under high performance requirements, and solve the problems of long design cycle, low efficiency and high cost of traditional TiAl-based alloy new materials.

[0099] In order to achieve faster and more accurate design of TiAl-based alloy composition, this invention optimizes the model based on the above-mentioned high-performance multi-element TiAl-based alloy design method based on machine learning, thereby further improving the model accuracy.

[0100] The high-performance multi-component TiAl-based alloy design method based on machine learning in this invention further includes the following steps:

[0101] S7. Using the above-mentioned optimal alloy composition, TiAl-based alloys were prepared, and the relevant performance tests of the prepared alloys were performed to obtain experimental data on alloy performance.

[0102] In an embodiment of the present invention, TiAl-based alloys are prepared using a powder-coated hot isostatic pressing method, with the following process parameters:

[0103] The sintering temperature is 1200-1300℃, the holding pressure is 140-180MPa, the holding time is 4-6h, the heating rate is 5-10℃ / min, and after cooling, the cladding is removed by wire EDM to obtain the desired alloy.

[0104] S8 adds the optimal alloy composition and experimental data to the dataset as a new set of data in the dataset; where the optimal alloy composition is used as input data and the experimental data is used as optimal output data.

[0105] S9. Repeat steps S1 to S4 to iteratively optimize the TiAl-based alloy composition-performance optimal model.

[0106] It is worth mentioning that the determination coefficient (R²) of the model can be calculated. 2 Mean absolute percentage error (MAPE), typically R 2 A value ≥0.9 and MAPE ≤10% indicate that the model accuracy meets the requirements.

[0107] The following detailed description of the high-performance multi-component TiAl-based alloy composition design method based on machine learning in this invention will be provided through specific embodiments.

[0108] Example 1:

[0109] 1) Establishing an initial dataset: Obtain TiAl-based alloy composition and density data from publicly available literature, experimental data, and production data to establish an initial dataset; wherein the Al content is 45-48 at.% in atomic percentage.

[0110] 2) Normalization: The content of each alloy element and the density data of the alloy in the initial dataset are normalized (0~1) to obtain the TiAl-based alloy composition-density dataset. The dataset is divided into a training set and a test set. The data in the training set accounts for 80% of the total data in the dataset, and the data in the test set accounts for 20% of the total data in the dataset.

[0111] 3) Establish the initial model of TiAl-based alloy composition-density GBDT: use the alloy element content in the training set as input data and the alloy density data as output data. Set the number of randomly selected alloy elements N=3, the maximum number of steps to 250, and set the hyperparameters n_estimator to 600, learning_rate to 0.05, max_depth to 3 and subsample to 1.

[0112] 4) The hyperparameters of the initial GBDT model were optimized using the K-Fold Cross Validation algorithm to obtain an improved GBDT model. Here, K = 10, and the optimization intervals for the GBDT algorithm hyperparameters n_estimator, learning_rate, max_depth, and subsample are 1–1000, 0.01–0.5, 1–10, and 0–1, respectively. The number of iterations n_iter is 400. Simultaneously, the coefficient of determination R0 of the models before and after the improvement can be compared. 2 Among them, R before improvement 2 =0.90, improved R 2 =0.95.

[0113] 5) The accuracy of the GBDT improved model was evaluated using a test set. Specifically, the TiAl-based alloy composition of each test set was used as input data to obtain the second predicted density data (corresponding to the second predicted alloy properties data). Then, the coefficient of determination R was calculated using this second predicted density data and the density data (corresponding to the second alloy properties data) in each test set. 2=0.94, MAPE=2.6% (satisfying the conditions of ≥0.90 and MAPE≤10%), this improved GBDT model is also considered the optimal GBDT model.

[0114] 6) For the five-element alloy composition of Ti (balance), Al, Cr, Nb and Sn, three levels were determined for each factor to carry out orthogonal design of alloy composition, resulting in multiple sets of preset alloy composition. The multiple sets of preset alloy composition were then input into the GBDT optimal model as input data to obtain the first predicted data of density (corresponding to the first predicted data / predicted value of alloy properties) corresponding to the multiple sets of preset alloy composition, as shown in Table 1.

[0115] 7) The optimal alloy composition was determined by screening multiple sets of first-prediction density data in Table 1, with the screening criterion being a density ≥99%.

[0116] 8) Using the above-mentioned optimal alloy composition, TiAl-based alloy parts were prepared by powder-coated hot isostatic pressing (POP). The specific process parameters were: sintering temperature of 1280℃, holding pressure of 160MPa, holding time of 5h, and heating rate of 5℃ / min. After cooling, the cladding was removed by wire electrical discharge machining to obtain TiAl-based alloy samples. Two samples were taken from the upper, middle, and lower parts of the sample, and the alloy density was tested by Archimedes' water displacement method. The average value (corresponding to experimental data / experimental value) was taken. The results are shown in Table 1.

[0117] 9) Compare the percentage error between the predicted and experimental values, and add the experimental data to the established dataset. Continuously improve the model's prediction accuracy and generalization ability through iterative feedback of experimental data.

[0118] Table 1. Composition design of TiAl-based alloys and corresponding first predicted and experimental data on density.

[0119]

[0120]

[0121] As shown in Table 1, for the pentagonal TiAl-based alloy composition of Ti, Al, Cr, Nb, and Sn, the percentage error between the experimental and predicted values ​​of the density of powder hot isostatic pressing TiAl-based parts predicted by the machine learning model proposed in this invention is within the range of 0.05% to 0.41%. This indicates that the TiAl-based alloy composition-density GBDT model established in this invention has high accuracy (generally, an error within 10% indicates that the model accuracy meets the requirements, and the smaller the error, the higher the accuracy).

[0122] The optimal alloy composition in Example 1 is: Al 46-48 at.%, Cr 0-2 at.%, Nb 0-2 at.%, Sn 1-2 at.%, with the balance being Ti.

[0123] Example 2:

[0124] 1) Establishing an initial dataset: Obtain TiAl-based alloy composition and room temperature mechanical properties (tensile strength and elongation) data from publicly available literature, experimental data, and production data to establish an initial dataset; wherein the Al content is 45-49 at.% in atomic percentage.

[0125] 2) Normalization: The content of each alloy element and the room temperature mechanical properties of the alloy in the initial dataset are normalized (0~1) to obtain the TiAl-based alloy composition-room temperature mechanical properties dataset. The dataset is divided into a training set and a test set. The data in the training set accounts for 70% of the total data in the dataset, and the data in the test set accounts for 30% of the total data in the dataset.

[0126] 3) Establish an initial SVR model for TiAl alloy composition-room temperature mechanical properties: The alloy element content in the training set is used as input data, and the room temperature tensile strength and room temperature elongation of the alloy are used as output data. The number of randomly selected alloy elements is set to N=4, the maximum number of steps is 400, and the hyperparameters are set to c=60, gamma=0.1 and epsilon=0.06.

[0127] 4) The hyperparameters of the initial SVR model are optimized using the K-Fold Cross Validation algorithm to obtain an improved SVR model. Here, K = 5, and the optimization ranges for the SVR algorithm hyperparameters c, gamma, and epsilon are 1–100, 0.001–1, and 0.001–1, respectively. The number of iterations n_iter is 500. Simultaneously, the coefficient of determination R of the models before and after the improvement can be compared. 2 Among them, R before improvement 2 =0.91, improved R 2 =0.94.

[0128] 5) The accuracy of the SVR improved model is evaluated using a test set. Specifically, the TiAl-based alloy composition of each set of test data is used as input data to obtain the second predicted data for room temperature mechanical properties (corresponding to the second predicted data for alloy properties). Then, the coefficient of determination R is calculated using this second predicted data for room temperature mechanical properties and the room temperature mechanical property data (corresponding to the second alloy property data) in each set of test data. 2 =0.93, MAPE=3.1% (satisfying the conditions of ≥0.90 and MAPE≤10%), this improved SVR model is also considered the optimal SVR model.

[0129] 6) The six-element alloy composition of Ti (balance), Al, Cr, Nb, B and V is designed by free combination to obtain multiple sets of preset alloy compositions. These multiple sets of preset alloy compositions are then used as input data into the SVR optimal model to obtain the first predicted data of room temperature mechanical properties corresponding to the multiple sets of preset alloy compositions (first predicted data / predicted value of corresponding alloy properties), as shown in Table 2.

[0130] 7) The optimal alloy composition was determined by screening multiple sets of room temperature mechanical properties first prediction data in Table 2. The screening criteria were tensile strength ≥ 520% ​​and elongation ≥ 1.5%.

[0131] 8) Using the above-mentioned optimal alloy composition, TiAl-based alloy parts were prepared by powder-coated hot isostatic pressing technology. The specific process parameters were: sintering temperature of 1260℃, holding pressure of 180MPa, holding time of 4h, and heating rate of 8℃ / min. After cooling, the cladding was removed by wire electrical discharge machining to obtain TiAl-based alloy samples. One M6 standard tensile specimen was taken from the center and one from the edge of the sample cross-section for room temperature tensile testing. The tensile strength and elongation were taken as average values ​​(corresponding to experimental data / experimental values). The results are shown in Table 2.

[0132] 9) Compare the percentage error between the predicted and experimental values, and add the experimental data to the established dataset. Continuously improve the model's prediction accuracy and generalization ability through iterative feedback of experimental data.

[0133] Table 2. TiAl-based alloy composition design and corresponding first-predicted and experimental data for room-temperature tensile strength and elongation.

[0134]

[0135]

[0136] As shown in Table 2, for the six-element TiAl-based alloy composition of Ti, Al, Cr, Nb, B, and V, the percentage error between the experimental and predicted values ​​of the tensile strength of the powder hot isostatic pressing TiAl-based alloy predicted by the machine learning model proposed in this invention is within the range of 1.44% to 3.95%, and the percentage error between the experimental and predicted values ​​of the elongation is within the range of 3.23% to 9.20%, both less than 10%. Therefore, the SVR model established by this invention based on the elitist strategy of genetic algorithm meets the accuracy requirements and can realize the rapid and accurate prediction of the room temperature tensile mechanical properties of multi-element TiAl-based alloys, thereby efficiently determining the optimal alloy composition and completing the accurate prediction and efficient design of TiAl-based alloy composition systems under high performance requirements.

[0137] The optimal alloy composition in Example 2 is: Al 45-47 at.%, Cr 1-2% at.%, Nb 5-8 at.%, B 0.1 at.%, V 1-2 at.%, with the balance being Ti.

[0138] This invention, based on the elitist strategy of genetic algorithms, combines random search and regression model algorithms, which ensures that the optimal parameters obtained in each iteration are retained in the final TiAl-based alloy composition-performance prediction model. This effectively solves the problem of poor prediction accuracy of multi-dimensional feature parameter models, enabling efficient and accurate prediction of TiAl-based alloy composition under high-performance requirements. It can significantly shorten the design cycle of multi-element TiAl-based alloys, improve alloy design efficiency, and reduce design costs.

[0139] It should be noted that the term "comprising" and any variations thereof in the specification and claims of this invention are intended to cover non-exclusive inclusion, for example, including a series of components that are not necessarily limited to those explicitly listed, but may include other components that are not explicitly listed or that are inherent to the component.

[0140] In this invention, the terms "upper," "lower," "bottom," "top," "left," "right," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing the invention and its embodiments, and are not intended to limit the indicated device, element, or component to having a specific orientation, or to be constructed and operated in a specific orientation.

[0141] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in certain situations to indicate a dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this invention based on the specific circumstances.

[0142] Furthermore, the descriptions of "first," "second," etc., involved in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features.

[0143] Furthermore, the technical solutions of the various embodiments can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0144] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A high-performance multi-component TiAl-based alloy composition design method based on machine learning, characterized in that, Includes the following steps: The composition and performance data of TiAl-based alloys were acquired to establish a dataset. This dataset is divided into a training set and a test set. The training set contains multiple sets of training data, each including the composition of a first alloy and its corresponding performance data. The test set contains multiple sets of test data, each including the composition of a second alloy and its corresponding performance data. The Al atomic percentage in the TiAl-based alloy composition is 45-49 at.%. The training set is used for model training to establish an initial TiAl-based alloy composition-performance model. The establishment of the initial model includes the following steps: randomly selecting N alloy elements from the first alloy composition of the training set to form an initial variable subset; assigning the initial variable subset to set G, and using it as the input subset to call a regression model algorithm for the first iteration to construct a regression model; after completing the first iteration, storing the input components appearing in the regression model in the variable subset, and discarding the input components not appearing in the regression model; randomly selecting N alloy elements from the first alloy composition of the training set again to form a new variable subset, and assigning it to set G again for the next iteration; repeating the iteration until the mean absolute percentage error (MAPE) no longer decreases or reaches the maximum number of steps, thus obtaining the initial model. The initial model was optimized using the K-Fold Cross Validation algorithm to obtain an improved model; The improved model is evaluated and optimized using the test set to obtain the best model. By inputting multiple sets of preset alloy compositions into the optimal model, the first predicted data of the multiple alloy properties corresponding to them are obtained. The optimal alloy composition is determined based on the first predicted data of the multi-combination alloy properties.

2. The high-performance multi-element TiAl-based alloy design method based on machine learning as described in claim 1, characterized in that, The regression model algorithm adopts any one of support vector regression, decision tree regression, or gradient boosting regression tree algorithm.

3. The high-performance multi-element TiAl-based alloy design method based on machine learning as described in claim 1, characterized in that, N takes values ​​from 3 to 5.

4. The high-performance multi-element TiAl-based alloy design method based on machine learning as described in claim 1, characterized in that, The evaluation and optimization of the improved model using the test set includes: The second alloy composition is input into the improved model to obtain the second predicted data of alloy properties; The determination coefficient R is calculated using the second predicted data of the alloy properties and the second alloy property data. 2 and mean absolute percentage error (MAPE); Coefficient of determination R 2 Whether it is greater than or equal to a preset threshold, and whether the mean absolute percentage error (MAPE) is less than or equal to a preset threshold: When the coefficient of determination R 2 If the mean absolute percentage error (MAPE) is greater than or equal to the preset threshold and less than or equal to the preset threshold, the model optimization design is complete. When the coefficient of determination R 2 <Preset threshold or mean absolute percentage error (MAPE)> Preset threshold, repeat the optimization design steps until the coefficient of determination R is reached. 2 ≥Preset threshold and mean absolute percentage error (MAPE) ≤Preset threshold.

5. The high-performance multi-element TiAl-based alloy design method based on machine learning as described in claim 1, characterized in that, Also includes: TiAl-based alloys were prepared using the optimal alloy composition, and experimental data on alloy properties were obtained. The optimal alloy composition and the experimental data are added to the dataset to form a new set of data. The optimal model is then iteratively optimized.

6. The high-performance multi-element TiAl-based alloy design method based on machine learning as described in claim 5, characterized in that, TiAl-based alloys were prepared using a powder-coated hot isostatic pressing method, with the following process parameters: The sintering temperature is 1200~1300℃, the holding pressure is 140~180MPa, the holding time is 4~6h, and the heating rate is 5~10℃ / min.

7. The high-performance multi-element TiAl-based alloy design method based on machine learning as described in claim 1, characterized in that, The training set contains 70-80% of the total data in the dataset, and the test set contains 20-30% of the total data in the dataset. The sum of the ratios of the training set and the test set is always 1.

8. The high-performance multi-element TiAl-based alloy design method based on machine learning as described in claim 1, characterized in that, The TiAl-based alloy composition includes Ti, Al, and at least one of Cr, Nb, Mo, W, V, Ta, C, N, Ni, Co, Sn, Si, and Cu.

9. The high-performance multi-element TiAl-based alloy design method based on machine learning as described in claim 1, characterized in that, The properties of the TiAl-based alloy include at least one of the following: density, room temperature tensile strength, room temperature elongation, 750℃ tensile strength, 750℃ elongation, and high-temperature oxidation weight gain rate.

Images